feat: 收视分析看板前端 L1-L4 实现 + 25期真实数据导入

收视分析页面完整实现：指标卡（含四档动画）、走势折线图（dataZoom滑块+确认按钮）、
季度/编导/题材对比（双列布局）、双引擎象限图（题材热度×叙事结构散点）。
导入25期真实收视数据及AI标签，修复侧边栏fixed定位和滚轮冲突。

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

scripts/fix_label_mapping.py

@@ -0,0 +1,106 @@
+"""
+修复 AI 标签映射：ground-truth ep编号 ≠ Excel 播出期号
+用标题模糊匹配重新关联正确的 AI 标签
+"""
+
+import json
+import sys
+import os
+from difflib import SequenceMatcher
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), '..', 'backend'))
+
+from app.db.session import get_session
+from app.models.episode import Episode
+from app.models.user import User
+from sqlmodel import select
+
+project_root = os.path.join(os.path.dirname(__file__), '..')
+
+# 读 ground-truth
+with open(os.path.join(project_root, 'ai-labeling', 'benchmark-set', 'ground-truth.json'), 'r', encoding='utf-8') as f:
+    gt_data = json.load(f)
+
+gt_episodes = gt_data['episodes']
+
+# 读数据库
+session = next(get_session())
+db_episodes = session.exec(select(Episode).order_by(Episode.episode_number)).all()
+
+print(f"DB episodes: {len(db_episodes)}")
+print(f"GT episodes: {len(gt_episodes)}")
+print()
+
+# ── 标题相似度匹配 ──
+def title_similarity(a, b):
+    """计算两个标题的相似度"""
+    # 清理标题
+    clean_a = a.replace('"', '').replace('"', '').replace("'", '').replace('——', '').replace('：', '').replace(' ', '').replace('《', '').replace('》', '')
+    clean_b = b.replace('"', '').replace('"', '').replace("'", '').replace('——', '').replace('：', '').replace(' ', '').replace('《', '').replace('》', '')
+    return SequenceMatcher(None, clean_a, clean_b).ratio()
+
+# 对每个 DB episode，找最佳匹配的 GT episode
+matched = []
+used_gt = set()
+
+for db_ep in db_episodes:
+    best_score = 0
+    best_gt = None
+
+    for i, gt_ep in enumerate(gt_episodes):
+        if i in used_gt:
+            continue
+
+        gt_title = gt_ep.get('title', '')
+        score = title_similarity(db_ep.program_name, gt_title)
+
+        # 也检查份额是否匹配（辅助判断）
+        gt_share = gt_ep.get('share')
+        share_match = (gt_share is not None and db_ep.audience_share is not None
+                      and abs(float(gt_share) - float(db_ep.audience_share)) < 0.01)
+
+        # 份额匹配加分
+        if share_match:
+            score += 0.3
+
+        if score > best_score:
+            best_score = score
+            best_gt = (i, gt_ep)
+
+    if best_gt and best_score > 0.3:
+        used_gt.add(best_gt[0])
+        matched.append((db_ep, best_gt[1], best_score))
+        status = 'OK' if best_score > 0.5 else 'WEAK'
+        print(f"[{status}] DB ep{db_ep.episode_number:02d} \"{db_ep.program_name[:15]}\" -> GT ep{best_gt[1]['ep']:02d} \"{best_gt[1]['title'][:15]}\" (score={best_score:.2f})")
+    else:
+        print(f"[MISS] DB ep{db_ep.episode_number:02d} \"{db_ep.program_name[:15]}\" -> NO MATCH")
+        matched.append((db_ep, None, 0))
+
+print(f"\nMatched: {sum(1 for _,gt,_ in matched if gt is not None)}/{len(db_episodes)}")
+
+# ── 确认后更新 ──
+print("\n--- Updating AI labels ---")
+updated = 0
+for db_ep, gt_ep, score in matched:
+    if gt_ep is None:
+        continue
+
+    db_ep.program_format = gt_ep.get('program_format')
+    db_ep.equipment_domain = gt_ep.get('equipment_domain')
+    db_ep.scene_tags = gt_ep.get('scene_tags')
+    db_ep.tech_tags = gt_ep.get('tech_tags')
+    db_ep.narrative_structure = gt_ep.get('narrative_structure')
+    db_ep.opening_hook = gt_ep.get('opening_hook')
+    db_ep.ai_label_confidence = 'reviewed'
+
+    session.add(db_ep)
+    updated += 1
+
+session.commit()
+print(f"Updated {updated} episodes")
+
+# ── 验证 ──
+print("\n--- Verification ---")
+db_episodes = session.exec(select(Episode).order_by(Episode.episode_number)).all()
+for ep in db_episodes:
+    print(f"ep{ep.episode_number:02d} | {ep.program_name[:18]:18s} | {ep.program_format or '-':8s} | {ep.narrative_structure or '-':6s} | {ep.opening_hook or '-'}")