fix(llm): RAG citation pages, scrub placeholders, disease-aware document choice

- converter marks a paragraph glued across a page break with ⟦стр. N⟧; the chunker
  strips the marker and extends the chunk's page range (wrong-page citations 24 -> 3,
  the rest are scale-form table rows split by a page)
- [NAME]/[PHONE] placeholders from privacy.scrub() no longer count as unknown words
  (a scrubbed question used to fall below the coverage gate)
- a question naming one disease boosts its guideline in retrieval and passage choice
  ("ТМС при депрессии" was answered from the PTSD guideline); "антидепрессант" is not a cue

Refs #62

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

app/llm/rag/qa.py

@@ -198,6 +198,7 @@ def best_passage(
     audience: Audience,
     cfg: AnswerConfig,
     defined: set[str] | None = None,
+    cued: set[str] | None = None,
 ) -> Passage | None:
     """Best sentence of the chunk extended by adjacent relevant sentences (contiguous, within limits)."""
     sents = [s if _allowed(s, audience) else None for s in chunk_sentences(chunk, clinical=audience != "patient")]
@@ -235,7 +236,7 @@ def best_passage(
         size += length(nxt)
     text = " ".join(s for s in sents[lo : hi + 1] if s)
     definition = scores[best] > index.overlap(head, terms)  # got DEFINITION_BONUS
-    return Passage(chunk, text, index.overlap(text, terms) * index.boost(chunk, audience), definition)
+    return Passage(chunk, text, index.overlap(text, terms) * index.boost(chunk, audience, cued), definition)
 
 
 def select(index: BM25Index, question: str, audience: Audience, docs: set[str] | None = None) -> list[Passage]:
@@ -244,7 +245,8 @@ def select(index: BM25Index, question: str, audience: Audience, docs: set[str] |
     hits = index.search(question, audience=audience, docs=docs, top_k=cfg.max_chunks)
     terms = index.query_terms(question)
     defined = _defined_term(question)
-    passages = [p for h in hits if (p := best_passage(index, h.chunk, terms, audience, cfg, defined))]
+    cued = index.cued_docs(question)
+    passages = [p for h in hits if (p := best_passage(index, h.chunk, terms, audience, cfg, defined, cued))]
     if not passages:
         return []
     # A definition answers "что такое X?" better than any passage that merely mentions X often.