fix(backend): 落实 PR #12 评审意见

- P1 事件循环让出：run_rag 在样本边界 await asyncio.sleep(0)，运行中取消/进度/SSE 可及时调度
- P2 SSE 终止事件：历史回放期间识别终止事件并结束流，try/finally 保证订阅清理
- P2 FTS 截断：fts 走数据库侧精确分页与计数，阈值经 bm25 截止值换算，不再受 5000 条固定截断
- P2 仅块标注：expected_block_ids 从块反查所属笔记，避免合法样本被判零分

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

backend/app/retrieval/engine.py

@@ -29,8 +29,6 @@ from app.textutils import make_snippet, match_query
 CANDIDATE_POOL = 50
 # 分页窗口上限：候选池至少覆盖 offset+limit，但设上限防止超大 offset 撑爆内存
 MAX_CANDIDATE_POOL = 200
-# FTS 全量取回上限：统一归一化 + 阈值过滤后再分页，保证阈值语义跨页一致
-FTS_FETCH_LIMIT = 5000
 # 带 metadata 过滤时放大召回倍数，缓解「先截断候选池再过滤」造成的漏召回
 OVERSCAN_FACTOR = 4
 
@@ -58,7 +56,7 @@ class RetrievalEngine:
         # 候选池至少覆盖本次请求的 offset+limit，保证分页能取到目标页；设上限防内存失控
         window = min(request.offset + request.limit, MAX_CANDIDATE_POOL)
         pool_size = max(CANDIDATE_POOL, window)
-        # 带过滤时放大召回；FTS 则一次性取全量命中（≤FTS_FETCH_LIMIT）避免截断漏召回
+        # 带过滤时放大召回，缓解「先截断候选池再过滤」造成的漏召回
         recall = min(pool_size * OVERSCAN_FACTOR, MAX_CANDIDATE_POOL) if has_filters else pool_size
 
         # 1. 按模式收集候选（FTS 与 Vector 各产出「按相关性降序」的 block_id 列表）
@@ -126,7 +124,7 @@ class RetrievalEngine:
         # score_threshold：归一化后过滤低分结果（默认 0 不过滤）
         ordered = [(bid, score) for bid, score in ordered if score >= request.score_threshold]
 
-        # 5. 分页：total = 过滤后候选集大小。fts 已取全量（≤FTS_FETCH_LIMIT）故为真实命中数；
+        # 5. 分页：total = 过滤后候选集大小。fts 走数据库精确分页，total 为真实命中数；
         #    vector/hybrid 为 KNN 候选集，无全局 total。
         total = len(ordered)
         page = ordered[request.offset : request.offset + request.limit]
@@ -139,18 +137,18 @@ class RetrievalEngine:
         )
 
     def _search_fts(self, request: SearchRequest) -> SearchResponse:
-        """FTS 专用路径：先取全量命中（≤FTS_FETCH_LIMIT），统一归一化 + 阈值过滤后再分页。
+        """FTS 专用路径：在数据库侧完成过滤、计数与分页，不取全量后再截断。
 
-        阈值过滤必须在计数与分页之前完成，否则 score_threshold 只作用于当前页，
-        且返回的 total 与 items 数量不一致（如 items 为空但 total 非零）。"""
+        阈值过滤时，min-max 归一化是 bm25 的线性函数，据此把 score_threshold 换算为
+        bm25 截止值（bm25_max），使过滤、计数与分页口径一致；无阈值时走数据库原生分页，
+        total 始终为过滤后的真实命中数，不再受固定截断影响。
+        """
         match = match_query(request.query)
         if not match:
             return self._empty(request)
 
-        fts_hits, _ = repository.fts_search_page(
+        bounds = repository.fts_score_bounds(
             match=match,
-            limit=FTS_FETCH_LIMIT,
-            offset=0,
             folders=request.folders,
             note_ids=request.note_ids,
             tags=request.tags,
@@ -159,17 +157,47 @@ class RetrievalEngine:
             updated_from=request.updated_from,
             updated_to=request.updated_to,
         )
-        if not fts_hits:
+        if bounds is None:
             return self._empty(request)
 
-        ordered = normalize_scores([(hit.block_id, -hit.bm25) for hit in fts_hits])
-        ordered = [(bid, score) for bid, score in ordered if score >= request.score_threshold]
-        total = len(ordered)
-        page = ordered[request.offset : request.offset + request.limit]
-        hits = {h.block_id: h for h in repository.get_block_hits([bid for bid, _ in page])}
+        lo, hi = bounds
+        bm25_max: float | None = None
+        if request.score_threshold > 0:
+            # norm = (hi - bm25) / (hi - lo)；norm >= threshold ⟺ bm25 <= hi - threshold*(hi - lo)
+            bm25_max = hi - request.score_threshold * (hi - lo)
+
+        fts_hits, total = repository.fts_search_page(
+            match=match,
+            limit=request.limit,
+            offset=request.offset,
+            folders=request.folders,
+            note_ids=request.note_ids,
+            tags=request.tags,
+            created_from=request.created_from,
+            created_to=request.created_to,
+            updated_from=request.updated_from,
+            updated_to=request.updated_to,
+            bm25_max=bm25_max,
+        )
+        if not fts_hits:
+            # 本页无结果：offset 越过末页时 total 仍为真实命中数（>0），需保留而非归零
+            return SearchResponse(
+                query=request.query,
+                mode=request.mode,
+                items=[],
+                page=PageMeta(total=total, limit=request.limit, offset=request.offset),
+            )
+
+        # 分数按全局 bm25 上下界归一化（与取全量后 normalize_scores 等价），保证跨页一致
+        span = hi - lo
+        if span == 0:
+            ordered = [(hit.block_id, 1.0) for hit in fts_hits]
+        else:
+            ordered = [(hit.block_id, round((hi - hit.bm25) / span, 6)) for hit in fts_hits]
+        hits = {h.block_id: h for h in repository.get_block_hits([bid for bid, _ in ordered])}
         items = [
             self._build_result(hits[block_id], request, score)
-            for block_id, score in page
+            for block_id, score in ordered
             if block_id in hits
         ]
         return SearchResponse(