feat: 事件时间轴缩略帧 + 人物管理头像 + 人物合并硬规则校验

## 新架构：Oracle 集中计算 + NAS 代理展示

### Oracle 端 (fam-edge)
- 新增 frame_service: ffmpeg 视频抽帧 + VLM 人物定位裁剪头像（磁盘缓存）
- 新增 /api/oracle/frame: 按 video_id+ts 抽帧返回 jpeg（带 token）
- 新增 /api/oracle/avatar: 按 label 生成人物头像（VLM 定位人物 + 兜底整帧居中）
- 新增 person_identifier: 人物身份识别模块
- Gemini 适配器支持 flash/flash-lite 双模型切换，429 自动降级
- frame_service VLM 全模型 429 时进入 10 分钟熔断，避免每次请求白打配额
- 兜底头像不落缓存，配额恢复后自动重试 VLM 精确定位

### 人物合并硬规则校验（框架级修复）
- person_service: LLM 合并结果落库前加硬冲突检测
  - 性别冲突 → 绝不合并
  - 年龄档跨未成年/成年 → 绝不合并（防止把爷爷/宝宝并进同一人）
- oracle_db: upsert_person 入口剥离括号后缀（人物A（别名：人物B） → 人物A），消灭垃圾人物行
- 修复 set_canonical 丢弃 source 参数的 bug（旧代码硬编码 'manual' 导致错误合并被永久固化）
- get_events_for_label: 只提取该身份组的特征文本，头像定位更精准

### NAS 端 (fam-core)
- 新增 img_proxy: /api/proxy/frame 和 /api/proxy/avatar 代理 Oracle 图片
- app.py 注册 img_bp 蓝图
- oracle_sync / db_layer / member_manager 同步人物表

### UI 端 (fam-ui)
- 事件时间轴: 每条事件卡片加时间点缩略帧
- 人物管理: 每人卡片加头像（150x150 圆角）
- parse_persons: 剥离括号备注，与 Oracle 归一化一致
- 新增 EventItem 组件、Timeline 页改造
- Chat / ServiceStatus 页相应调整

### 数据库
- scripts/ddl.sql: 同步表结构更新
- Oracle people 表: features_json / display_uid / source 字段完善

fam-edge/tests/test_oracle_db.py

@@ -181,3 +181,133 @@ def test_has_motion_in_range_local_filters_by_camera_id(tmp_path):
         [{"event_id": 1, "camera_id": 99, "event_type": 10, "start_time": 2500, "duration": 5}])
     assert db.has_motion_in_range_local(2000, 3000, camera_id=2) is False
     assert db.has_motion_in_range_local(2000, 3000, camera_id=99) is True
+
+
+# ----------------------------------------------------------------------
+# 人物对应关系表（video_id, raw_uid) -> canonical_name
+# ----------------------------------------------------------------------
+
+def _seed_video_with_events(db, filename="motion_1_1000.mp4"):
+    vid = db.ensure_video(filename, f"/tmp/{filename}", event_start_time="2026-08-22 10:00:00")
+    events = [
+        {"timestamp": "10:00:01", "description": "在客厅走动", "people": ["人物A"],
+         "person_appearances": [{"uid": "人物A", "features": {"gender": "男"}, "action": "走动"}]},
+        {"timestamp": "10:00:05", "description": "坐下", "people": ["人物A", "人物B"],
+         "person_appearances": [
+             {"uid": "人物A", "features": {"gender": "男"}, "action": "坐下"},
+             {"uid": "人物B", "features": {"gender": "女"}, "action": "站立"}]},
+    ]
+    db.mark_video_processed(vid, "摘要", events, ["人物A", "人物B"], "gemini")
+    return vid
+
+
+def test_set_identity_mapping_inserts_new_row(tmp_path):
+    db = _db(tmp_path)
+    assert db.set_identity_mapping(1, "人物A", "爷爷", source="auto_id") is True
+    assert db.get_identity_map_for_video(1) == {"人物A": "爷爷"}
+
+
+def test_set_identity_mapping_updates_existing_non_manual_row(tmp_path):
+    db = _db(tmp_path)
+    db.set_identity_mapping(1, "人物A", "爷爷", source="auto_id")
+    assert db.set_identity_mapping(1, "人物A", "爸爸", source="auto_id") is True
+    assert db.get_identity_map_for_video(1) == {"人物A": "爸爸"}
+
+
+def test_set_identity_mapping_manual_protected_from_auto_overwrite(tmp_path):
+    """核心诉求: 人工纠正过的映射不能被后续自动识别悄悄改回去。"""
+    db = _db(tmp_path)
+    db.set_identity_mapping(1, "人物A", "爸爸", source="manual")
+    changed = db.set_identity_mapping(1, "人物A", "爷爷", source="auto_id")
+    assert changed is False
+    assert db.get_identity_map_for_video(1) == {"人物A": "爸爸"}
+
+
+def test_set_identity_mapping_manual_can_override_manual(tmp_path):
+    db = _db(tmp_path)
+    db.set_identity_mapping(1, "人物A", "爸爸", source="manual")
+    changed = db.set_identity_mapping(1, "人物A", "爷爷", source="manual")
+    assert changed is True
+    assert db.get_identity_map_for_video(1) == {"人物A": "爷爷"}
+
+
+def test_set_identity_mapping_no_change_returns_false(tmp_path):
+    db = _db(tmp_path)
+    db.set_identity_mapping(1, "人物A", "爷爷", source="auto_id")
+    changed = db.set_identity_mapping(1, "人物A", "爷爷", source="auto_id")
+    assert changed is False
+
+
+def test_get_identity_map_for_video_scoped_per_video(tmp_path):
+    """核心诉求: 同一个 raw_uid 字符串在不同视频里可能是不同真人，映射必须按
+    video_id 隔离，不能串。"""
+    db = _db(tmp_path)
+    db.set_identity_mapping(1, "人物A", "爷爷", source="auto_id")
+    db.set_identity_mapping(2, "人物A", "爸爸", source="auto_id")
+    assert db.get_identity_map_for_video(1) == {"人物A": "爷爷"}
+    assert db.get_identity_map_for_video(2) == {"人物A": "爸爸"}
+
+
+def test_rewrite_event_person_names_updates_events_and_video(tmp_path):
+    db = _db(tmp_path)
+    vid = _seed_video_with_events(db)
+    db.rewrite_event_person_names(vid, {"人物A": "爷爷", "人物B": "媳妇"})
+
+    rows = db._conn.execute(
+        "SELECT person_list_json, person_appearances_json FROM events "
+        "WHERE video_id=? ORDER BY id", (vid,)).fetchall()
+    assert json.loads(rows[0]["person_list_json"]) == ["爷爷"]
+    pa0 = json.loads(rows[0]["person_appearances_json"])
+    assert pa0[0]["uid"] == "爷爷"
+    assert json.loads(rows[1]["person_list_json"]) == ["爷爷", "媳妇"]
+    pa1 = json.loads(rows[1]["person_appearances_json"])
+    assert {p["uid"] for p in pa1} == {"爷爷", "媳妇"}
+
+    vrow = db._conn.execute("SELECT people_json FROM videos WHERE id=?", (vid,)).fetchone()
+    assert set(json.loads(vrow["people_json"])) == {"爷爷", "媳妇"}
+
+
+def test_rewrite_event_person_names_noop_on_empty_map(tmp_path):
+    db = _db(tmp_path)
+    vid = _seed_video_with_events(db)
+    before = db._conn.execute(
+        "SELECT person_list_json FROM events WHERE video_id=?", (vid,)).fetchall()
+    db.rewrite_event_person_names(vid, {})
+    after = db._conn.execute(
+        "SELECT person_list_json FROM events WHERE video_id=?", (vid,)).fetchall()
+    assert [r["person_list_json"] for r in before] == [r["person_list_json"] for r in after]
+
+
+def test_correct_video_identity_end_to_end(tmp_path):
+    """核心诉求: 纠错入口应该找到当前展示名对应的映射行，改写映射 + 立即重写
+    展示数据，且标记为 manual（受保护）。"""
+    db = _db(tmp_path)
+    vid = _seed_video_with_events(db)
+    db.set_identity_mapping(vid, "人物A", "爷爷", source="auto_id")
+    db.rewrite_event_person_names(vid, {"人物A": "爷爷"})
+
+    db.correct_video_identity(vid, current_name="爷爷", new_name="爸爸")
+
+    assert db.get_identity_map_for_video(vid) == {"人物A": "爸爸"}
+    rows = db._conn.execute(
+        "SELECT person_list_json FROM events WHERE video_id=? ORDER BY id", (vid,)).fetchall()
+    assert json.loads(rows[0]["person_list_json"]) == ["爸爸"]
+    # manual 之后不能被自动识别覆盖回去
+    changed = db.set_identity_mapping(vid, "人物A", "爷爷", source="auto_id")
+    assert changed is False
+
+
+def test_correct_video_identity_without_prior_mapping_uses_current_name_as_raw_uid(tmp_path):
+    """核心诉求: 老流水线时代产出的数据从没跑过闭集识别，映射表里没有记录——
+    纠错依然要能生效，把 current_name 本身当 raw_uid 存一条新映射。"""
+    db = _db(tmp_path)
+    vid = db.ensure_video("motion_2_2000.mp4", "/tmp/x.mp4", event_start_time="2026-08-22 10:00:00")
+    events = [{"timestamp": "10:00:01", "description": "走动", "people": ["爷爷"],
+              "person_appearances": [{"uid": "爷爷", "features": {"gender": "男"}, "action": "走动"}]}]
+    db.mark_video_processed(vid, "摘要", events, ["爷爷"], "gemini")
+
+    db.correct_video_identity(vid, current_name="爷爷", new_name="爸爸")
+    assert db.get_identity_map_for_video(vid) == {"爷爷": "爸爸"}
+    rows = db._conn.execute(
+        "SELECT person_list_json FROM events WHERE video_id=?", (vid,)).fetchall()
+    assert json.loads(rows[0]["person_list_json"]) == ["爸爸"]