fix(memory): hybrid_recall log 把 passed 数挂在 thresh 字段误导调参 (#1413)

`(>thresh %d)` 看着是阈值，实际填的是 len(bm25_top) / len(cosine_top) —— 过
阈值后剩下的数量。线上看到 `scored bm25=0 (>thresh 0)` 会被读成"阈值就是 0
所以全过"，实际语义是"打分后只剩 0 篇，过阈值的也是 0 篇"，阈值常量
HYBRID_RECALL_BM25_THRESHOLD=0.1 完全没出现在 log 里。

拆成 `passed` + `thresh` 两段，过阈值数和阈值常量都显式标出来：

  scored bm25=72 (passed 5, thresh=0.10) emb=43 (passed 4, thresh=4.00)

Co-authored-by: Hongzhi Wen <cartabio.coder1@gmail.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: 3 people

memory/hybrid_recall.py

@@ -545,13 +545,17 @@ async def hybrid_recall(
 
     elapsed_ms = (time.time() - start) * 1000.0
     # union pool size for observability — bm25 pool is the superset.
+    # `passed` = items surviving the per-side threshold; `thresh` is the
+    # cutoff constant. 历史上这条 log 把 `passed` 数挂在 `(>thresh %d)`
+    # 字段里，被读成"阈值=N"误导调参，所以拆成 passed + thresh 两段。
     logger.info(
-        "[hybrid_recall] %s: pool bm25=%d emb=%d | scored bm25=%d (>thresh %d) "
-        "emb=%d (>thresh %d) | fused=%d | %.0fms",
+        "[hybrid_recall] %s: pool bm25=%d emb=%d | "
+        "scored bm25=%d (passed %d, thresh=%.2f) "
+        "emb=%d (passed %d, thresh=%.2f) | fused=%d | %.0fms",
         lanlan_name,
         len(bm25_pool), len(embedding_pool),
-        len(bm25_scored), len(bm25_top),
-        len(cosine_scored), len(cosine_top),
+        len(bm25_scored), len(bm25_top), HYBRID_RECALL_BM25_THRESHOLD,
+        len(cosine_scored), len(cosine_top), HYBRID_RECALL_COSINE_THRESHOLD,
         len(results), elapsed_ms,
     )
     return {