fix(douyin): 修复抖音真实链接503，增加第三方解析API兜底

根因：
- yt-dlp DouyinIE 当前需要 fresh cookies（即使只提取元信息），
  staging 上的 douyin_cookies.txt 是占位文件（3行），导致直链拿不到
- 原代码路径A（MediaKit ASR）因 direct_url=None 完全跳过
- 路径B（下载+本地ASR）同样因 cookies 失败返回友好 503

修复：
1. 修复 MediaKitClient monkey-patch 属性名错误
   （原用 self.base_url/self.api_key/self.timeout，实际是 self._base_url 等私有属性；
    错误添加了 _mk_headers 与类自带 _headers() 重复）
2. yt-dlp 增加 extractor_args 和 Accept-Language header
3. 新增第三方无水印解析 API 兜底（apizero.cn），
   yt-dlp 失败时自动切换，拿到直链后仍走 MediaKit ASR
4. 路径B 支持直接用第三方直链下载 MP4（不依赖 yt-dlp/cookies）
5. MediaKit ASR 轮询增加瞬时网络错误重试
6. 全部路径失败时返回友好 503（不再返回空 text 导致前端异常）

验证：本地通过 apizero 成功解析 https://v.douyin.com/hb-giW8cC1Q/
返回直链可下载（23MB mp4），HTTP 200

apps/api/app/api/routes/scripts_ai.py

@@ -179,6 +179,10 @@ def _extract_and_validate_douyin_url(raw_input):
 
 
 # ── MediaKitClient ASR 扩展（monkey patch） ────────────────────────────
+#
+# app/services/mediakit_client.py 已提供：
+#   self._api_key / self._base_url / self._timeout / self._headers() / self.is_available
+# 这里只补两个通用 JSON helper 和 ASR 提交/轮询方法。
 
 
 def _mk_post_json(self, path, payload):
@@ -200,9 +204,13 @@ def _mk_post_json(self, path, payload):
         ) from exc
     except httpx.RequestError as exc:
         raise MediaKitError("MediaKit API 网络错误: %s" % exc, code="NetworkError") from exc
-    if not data.get("success", True) and data.get("error"):
-        err = data["error"]
-        raise MediaKitError(err.get("message", "请求失败"), code=err.get("code", "RequestFailed"))
+    # 同步错误（提交参数错误等）：success=false 且含 error
+    if data.get("success") is False and data.get("error"):
+        err = data["error"] if isinstance(data["error"], dict) else {"message": str(data["error"])}
+        raise MediaKitError(
+            err.get("message", "请求失败"),
+            code=err.get("code", "RequestFailed"),
+        )
     return data
 
 
@@ -228,20 +236,29 @@ def _mk_get_json(self, path):
 
 
 def _mediakit_asr_submit(self, video_url):
+    """提交语音转字幕任务（POST /tools/asr-subtitles）。返回 task_id。"""
     data = self._post_json(
         "/tools/asr-subtitles",
         {"video_url": video_url, "language": "cmn-Hans-CN"},
     )
     task_id = data.get("task_id")
     if not task_id:
-        raise MediaKitError("MediaKit ASR 提交响应缺少 task_id")
+        raise MediaKitError("MediaKit ASR 提交响应缺少 task_id: %s" % str(data)[:200])
     return task_id
 
 
 def _mediakit_asr_poll(self, task_id, poll_interval=2.0, max_attempts=90):
-    for _ in range(max_attempts):
+    """轮询 ASR 任务直到 completed/failed。返回 (text, duration)。"""
+    for attempt in range(max_attempts):
         time.sleep(poll_interval)
-        data = self._get_json("/tasks/" + task_id)
+        try:
+            data = self._get_json("/tasks/" + task_id)
+        except MediaKitError as exc:
+            # 瞬时网络/超时可重试
+            if attempt < max_attempts - 1 and getattr(exc, "code", "") in ("Timeout", "NetworkError"):
+                logger.warning("MediaKit ASR 轮询异常（第%d次），将重试: %s", attempt + 1, exc)
+                continue
+            raise
         st = data.get("status")
         if st in ("completed", "success"):
             result = data.get("result") or {}
@@ -250,17 +267,23 @@ def _mediakit_asr_poll(self, task_id, poll_interval=2.0, max_attempts=90):
             duration = float(result.get("duration") or 0.0)
             return text.strip(), duration
         if st == "failed":
-            err = data.get("error") or {}
-            raise MediaKitError(
-                "MediaKit ASR 任务失败: %s" % err.get("message", "unknown"),
-                code=err.get("code", "TaskFailed"),
-            )
+            err = data.get("error")
+            if isinstance(err, dict):
+                msg = err.get("message") or "unknown"
+                code = err.get("code") or "TaskFailed"
+            elif isinstance(err, str):
+                msg, code = err, "TaskFailed"
+            else:
+                msg, code = "unknown", "TaskFailed"
+            raise MediaKitError("MediaKit ASR 任务失败: %s" % msg, code=code)
+        # running/pending/queued: continue
     raise MediaKitError(
         "MediaKit ASR 超时（%ss 未完成）" % int(poll_interval * max_attempts),
         code="Timeout",
     )
 
 
+# 绑定到类（零侵入，不修改原 mediakit_client.py）
 if not hasattr(MediaKitClient, "_post_json"):
     MediaKitClient._post_json = _mk_post_json
 if not hasattr(MediaKitClient, "_get_json"):
@@ -285,6 +308,7 @@ def _ytdlp_extract_video_url(page_url, cookiefile=None):
         "no_warnings": True,
         "noplaylist": True,
         "skip_download": True,
+        "extractor_args": {"douyin": {"webpage_cookie": ""}},  # 强制使用页面 cookie 流程
         "http_headers": {
             "User-Agent": (
                 "Mozilla/5.0 (Windows NT 10.0; Win64; x64) "
@@ -292,6 +316,7 @@ def _ytdlp_extract_video_url(page_url, cookiefile=None):
                 "Chrome/128.0.0.0 Safari/537.36"
             ),
             "Referer": "https://www.douyin.com/",
+            "Accept-Language": "zh-CN,zh;q=0.9,en;q=0.8",
         },
     }
     if cookiefile:
@@ -318,6 +343,125 @@ def _ytdlp_extract_video_url(page_url, cookiefile=None):
     return video_url, duration
 
 
+_PARSER_APIS = [
+    # 第三方抖音无水印解析 API（cookies 过期/yt-dlp 反爬升级时的兜底）
+    ("https://v1.apizero.cn/api/video-parse?url={url}&flat=1", "apizero"),
+]
+
+
+def _parser_extract_direct_url(page_url, timeout=10):
+    """使用第三方解析 API 获取抖音无水印直链（yt-dlp 失败时的兜底）。
+    返回 (direct_url, duration) 或 (None, 0.0)。
+    """
+    import urllib.parse
+    import httpx
+
+    for api_tpl, name in _PARSER_APIS:
+        url = api_tpl.format(url=urllib.parse.quote(page_url, safe=""))
+        try:
+            with httpx.Client(timeout=timeout, follow_redirects=True) as http:
+                resp = http.get(
+                    url,
+                    headers={
+                        "User-Agent": (
+                            "Mozilla/5.0 (Windows NT 10.0; Win64; x64) "
+                            "AppleWebKit/537.36 (KHTML, like Gecko) "
+                            "Chrome/128.0.0.0 Safari/537.36"
+                        ),
+                        "Accept": "application/json",
+                    },
+                )
+                resp.raise_for_status()
+                data = resp.json()
+        except Exception as exc:  # noqa: BLE001
+            logger.warning("第三方解析 %s 请求失败: %s", name, exc)
+            continue
+
+        try:
+            if data.get("code") != 0:
+                logger.warning("第三方解析 %s 返回错误: %s", name, data.get("msg"))
+                continue
+            vd = data.get("data") or {}
+            vlist = vd.get("video_list") or []
+            if not vlist:
+                continue
+            chosen = None
+            for v in vlist:
+                if not v.get("url"):
+                    continue
+                if "原画" in str(v.get("quality", "")):
+                    chosen = v
+                    break
+            if chosen is None:
+                chosen = vlist[0]
+            direct_url = chosen.get("url")
+            duration = float(vd.get("duration") or 0.0)
+            if direct_url:
+                logger.info("第三方解析 %s 成功，直链长度=%d", name, len(direct_url))
+                return direct_url, duration
+        except Exception as exc:  # noqa: BLE001
+            logger.warning("第三方解析 %s 响应解析失败: %s", name, exc)
+            continue
+
+    logger.warning("所有第三方解析均失败")
+    return None, 0.0
+
+
+def _direct_url_download_and_local_asr(direct_url, page_url, temp_dir):
+    """通过第三方解析得到的直链直接下载 MP4，再做本地 ASR。
+    返回 (text, duration)。
+    """
+    import httpx
+    import os
+
+    video_path = os.path.join(temp_dir, "video.mp4")
+    try:
+        with httpx.Client(timeout=60, follow_redirects=True) as http:
+            with http.stream(
+                "GET",
+                direct_url,
+                headers={
+                    "User-Agent": (
+                        "Mozilla/5.0 (Windows NT 10.0; Win64; x64) "
+                        "AppleWebKit/537.36 (KHTML, like Gecko) "
+                        "Chrome/128.0.0.0 Safari/537.36"
+                    ),
+                    "Referer": "https://www.douyin.com/",
+                },
+            ) as resp:
+                resp.raise_for_status()
+                downloaded = 0
+                with open(video_path, "wb") as f:
+                    for chunk in resp.iter_bytes(chunk_size=65536):
+                        if chunk:
+                            f.write(chunk)
+                            downloaded += len(chunk)
+                if downloaded == 0:
+                    raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail="直链下载为空")
+    except HTTPException:
+        raise
+    except httpx.TimeoutException:
+        logger.warning("直链下载超时: %s", page_url)
+        raise HTTPException(status_code=status.HTTP_504_GATEWAY_TIMEOUT, detail="视频下载超时，请稍后重试")
+    except Exception as exc:  # noqa: BLE001
+        logger.exception("直链下载失败: url=%s err=%s", page_url, exc)
+        raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail="视频下载失败: " + str(exc)[:200]) from exc
+
+    try:
+        text = transcribe_to_text(video_path)
+        return text.strip(), 0.0
+    except ASRNotConfiguredError as exc:
+        raise HTTPException(status_code=status.HTTP_503_SERVICE_UNAVAILABLE, detail=str(exc)) from exc
+    except ASRTranscriptionError as exc:
+        raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail=str(exc)) from exc
+    except Exception as exc:
+        logger.exception("直链下载后 ASR 转写异常: path=%s", video_path)
+        raise HTTPException(
+            status_code=status.HTTP_502_BAD_GATEWAY,
+            detail="语音识别失败: " + str(exc)[:200],
+        ) from exc
+
+
 def _ytdlp_download_and_local_asr(page_url, temp_dir, cookiefile=None):
     try:
         import yt_dlp
@@ -332,6 +476,7 @@ def _ytdlp_download_and_local_asr(page_url, temp_dir, cookiefile=None):
         "quiet": True,
         "no_warnings": True,
         "noplaylist": True,
+        "extractor_args": {"douyin": {"webpage_cookie": ""}},
         "http_headers": {
             "User-Agent": (
                 "Mozilla/5.0 (Windows NT 10.0; Win64; x64) "
@@ -339,6 +484,7 @@ def _ytdlp_download_and_local_asr(page_url, temp_dir, cookiefile=None):
                 "Chrome/128.0.0.0 Safari/537.36"
             ),
             "Referer": "https://www.douyin.com/",
+            "Accept-Language": "zh-CN,zh;q=0.9,en;q=0.8",
         },
     }
     if cookiefile:
@@ -447,6 +593,12 @@ def extract_from_douyin(
 
     # 路径 A：yt-dlp 拿直链 + MediaKit 云端 ASR
     direct_url, meta_duration = _ytdlp_extract_video_url(page_url, cookiefile=cookiefile)
+    if not direct_url:
+        logger.info("yt-dlp 未拿到直链，尝试第三方无水印解析 API 兜底")
+        try:
+            direct_url, meta_duration = _parser_extract_direct_url(page_url)
+        except Exception as exc:  # noqa: BLE001
+            logger.warning("第三方解析兜底异常: %s", exc)
     if meta_duration:
         duration = meta_duration
     _dbg("direct_url", direct_url or "<none>")
@@ -469,12 +621,26 @@ def extract_from_douyin(
     if not text:
         _dbg("fallback", "download+local_asr")
         with tempfile.TemporaryDirectory(prefix="douyin_extract_") as temp_dir:
-            text, dl_duration = _ytdlp_download_and_local_asr(
-                page_url, temp_dir, cookiefile=cookiefile
-            )
+            # 若已有第三方解析拿到的直链，优先直接下载 MP4 给本地 ASR
+            if direct_url:
+                _dbg("fallback_via", "direct_url_download")
+                text, dl_duration = _direct_url_download_and_local_asr(
+                    direct_url, page_url, temp_dir
+                )
+            else:
+                text, dl_duration = _ytdlp_download_and_local_asr(
+                    page_url, temp_dir, cookiefile=cookiefile
+                )
             if dl_duration and not duration:
                 duration = dl_duration
 
+    if not text:
+        logger.warning("抖音文案提取全部失败: url=%s", page_url)
+        raise HTTPException(
+            status_code=status.HTTP_503_SERVICE_UNAVAILABLE,
+            detail="抖音链接解析暂时不可用，请稍后重试或手动输入文案",
+        )
+
     return ExtractFromDouyinResponse(
         text=text,
         duration_seconds=duration,