perf(search): offload cosine scoring off event loop; document best-effort feed

Drift-audit Group 9 (param-cliff / unbounded search work):

- semantic_search_notes: the O(rows) cosine-similarity scoring loop ran
  synchronously on the event loop, so every RAG injection / search stalled
  other requests proportional to the user's embedding count. Move the scoring
  into asyncio.to_thread (results unchanged). The deeper fix — bounding the
  candidate set via pgvector ORDER BY/LIMIT — is noted as separate infra work.
- _semantic_knowledge_search: documented the best-effort top-N semantics —
   is the capped candidate-window size (not the true match count),
  matches beyond the cap aren't page-reachable, and each page recomputes the
  full merge. Prevents the silent-truncation trap; cached ranked-id paging /
  pgvector is the fix if exhaustive pagination is ever required.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

src/fabledassistant/services/knowledge.py

@@ -140,6 +140,14 @@ async def _semantic_knowledge_search(
     Exact keyword matches always rank above semantic-only matches so that
     searching for a name like "Weston" surfaces the note with that title
     before conceptually related notes.
+
+    BEST-EFFORT TOP-N, not exhaustive pagination: the ranked candidate set is
+    capped (keyword limit*2 + up to ~200 semantic), so `total` is the size of
+    that window, NOT the true match count, and matches beyond the cap are not
+    reachable by paging. Each page also recomputes the full merge (O(corpus)
+    per page). Acceptable for an interactive "best results" feed; a cached
+    ranked-id list or pgvector ORDER BY/LIMIT is the fix if exhaustive,
+    cheap pagination is ever needed.
     """
     # 1. Keyword search — title and body ILIKE
     keyword_notes: list[Note] = []