feat(rag): add hybrid retrieval + language layer on top of the vector baseline

Stacked on the vanilla vector RAG. Re-introduces the retrieval-quality layer:
keyword (FTS5/BM25) indexing, reciprocal rank fusion, MMR diversification,
adaptive-k, term-coverage boosting, and multilingual (Polish/English)
no-answer detection, with the HybridRetriever app↔lib boundary and full test
coverage (keywordIndex, rankFusion, hybridRetrieval, pipeline integration).

Hardening carried over from review:
- Keyword index backfills once from the existing vectors when it is empty, so
  documents imported on the vector-only build become keyword-searchable after
  upgrade instead of staying invisible forever.
- Keyword-only hits must clear a low semantic-similarity bar when a query
  embedding is available, so a loose prefix match (false friend) no longer
  qualifies unconditionally; keyword-without-embeddings behaviour is preserved.
- Semantic gate recalibrated to 0.40 with a top-keep floor (0.25) for the single
  best candidate, matching the on-device eval.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: kfaracik

__tests__/VectorStoreContext.test.tsx

@@ -32,6 +32,10 @@ jest.mock('../utils/embeddingModelMigration', () => ({
   migrateEmbeddingModelIfNeeded: jest.fn().mockResolvedValue(undefined),
 }));
 
+jest.mock('../database/keywordIndex', () => ({
+  ensureKeywordIndex: jest.fn().mockResolvedValue(undefined),
+}));
+
 jest.mock('../utils/embeddingModel', () => ({
   isEmbeddingModelDownloaded: jest.fn().mockResolvedValue(true),
 }));