fix: 修复 AI Code Review 3 个阻塞级问题

1. KeyError 风险: c['color_histogram'] → c.get('color_histogram', [])
   防止存量数据缺少该字段时 Worker 崩溃

2. 直方图默认值不对称: 任一方无数据时统一返回 0.0
   旧逻辑: 已有视频无直方图→0.5, 新视频无直方图→0.0
   新逻辑: 任一方无数据→0.0（无法判定相似）

3. Bhattacharyya 输入已归一化（cv2.normalize 保证）
   compute_color_histogram 已用 cv2.normalize 处理，无需额外改动

测试：49/49 全绿

apps/worker/video_processing/dedup.py

@@ -367,11 +367,14 @@ class VideoDeduplicator:
 
         # Step 4: 加权融合
         phash_similarity = 1.0 - (median_distance / 64)
-        hist_similarity = (
-            self._compute_histogram_similarity(fingerprint.color_histograms, existing_histograms)
-            if existing_histograms
-            else 0.5  # 无直方图数据时给中间值（向后兼容）
-        )
+
+        # 直方图相似度：任一方无数据时统一返回 0.0（无法判定），避免不对称
+        if existing_histograms and fingerprint.color_histograms:
+            hist_similarity = self._compute_histogram_similarity(
+                fingerprint.color_histograms, existing_histograms
+            )
+        else:
+            hist_similarity = 0.0
         combined_score = self.PHASH_WEIGHT * phash_similarity + self.HISTOGRAM_WEIGHT * hist_similarity
 
         return {
@@ -461,7 +464,7 @@ class VideoDeduplicator:
             chunk_data = self._get_existing_chunks(existing.id, session)
             if chunk_data:
                 existing_phashes = [c["phash_binary"] for c in chunk_data]
-                existing_histograms = [c["color_histogram"] for c in chunk_data]
+                existing_histograms = [c.get("color_histogram", []) for c in chunk_data]
             else:
                 # 回退：从 JSON 字段读取（存量旧视频）
                 existing_phashes = ef.get("keyframe_phashes", [])
@@ -526,7 +529,7 @@ class VideoDeduplicator:
             chunk_data = self._get_existing_chunks(existing.id, session)
             if chunk_data:
                 existing_phashes = [c["phash_binary"] for c in chunk_data]
-                existing_histograms = [c["color_histogram"] for c in chunk_data]
+                existing_histograms = [c.get("color_histogram", []) for c in chunk_data]
             else:
                 existing_phashes = ef.get("keyframe_phashes", [])
                 existing_histograms = ef.get("color_histograms", [])
@@ -616,7 +619,7 @@ class VideoDeduplicator:
             chunk_data = self._get_existing_chunks(existing.id, session)
             if chunk_data:
                 existing_phashes = [c["phash_binary"] for c in chunk_data]
-                existing_histograms = [c["color_histogram"] for c in chunk_data]
+                existing_histograms = [c.get("color_histogram", []) for c in chunk_data]
             else:
                 existing_phashes = ef.get("keyframe_phashes", [])
                 existing_histograms = ef.get("color_histograms", [])

tests/unit/test_dedup_v2.py

@@ -352,24 +352,24 @@ class TestBackwardCompatibility:
     """无直方图数据时不崩溃。"""
 
     def test_no_histogram_fallback(self):
-        """已有视频无分片直方图 → hist_similarity 回退到 0.5，不崩溃。"""
+        """已有视频无分片直方图 → hist_similarity 回退到 0.0，不崩溃。"""
         d = VideoDeduplicator()
         phashes = ["0" * 16] * 10
         fp = _make_fingerprint(phashes, [[0.5] * 96] * 10)
         # existing_histograms 为空列表
         result = d._check_fusion_duplicate(fp, {}, phashes, [])
-        # 应该不崩溃，hist_similarity=0.5
+        # 应该不崩溃，hist_similarity=0.0（无数据时统一为0）
         if result:
-            assert result["_debug"]["hist_similarity"] == 0.5
+            assert result["_debug"]["hist_similarity"] == 0.0
 
     def test_no_histogram_combined_score(self):
-        """无直方图时的 combined_score = 0.7 * phash + 0.3 * 0.5。"""
+        """无直方图时的 combined_score = 0.7 * phash + 0.3 * 0.0。"""
         d = VideoDeduplicator()
         phashes = ["0" * 16] * 10  # 完全相同
         fp = _make_fingerprint(phashes, [[0.5] * 96] * 10)
         result = d._check_fusion_duplicate(fp, {}, phashes, [])
         if result:
-            expected = 0.7 * 1.0 + 0.3 * 0.5  # = 0.85
+            expected = 0.7 * 1.0 + 0.3 * 0.0  # = 0.7
             assert result["similarity"] == pytest.approx(expected)