feat(search): hybrid scoring calibration, AST context enrichment, and inspector redesign

[spelech]

Summary

This PR addresses search calibration, AST context enrichment, and Inspector UX improvements:

1. Dynamic Hybrid Calibration & Search Modes:

   • Resolved BM25 suppression where fixed 70/30 dense bias capped exact symbol/keyword matches at 0.30.
   • Introduced dynamic dense weight parameter dense_weight (alpha in [0.0, 1.0]) with a balanced 0.5 default.
   • Added support for explicit search_mode: hybrid, semantic (pure vector similarity), and lexical (pure BM25 sparse matching) across vector stores (Qdrant, Chroma, PGVector).
   • Extended vector search returns with decomposed scores (score, dense_score, sparse_score, dense_rank, sparse_rank).

2. AST Metadata Context Enrichment:

   • Enriched search results with structural AST context queried from SQLite ast_symbols.
   • Populated exact function/method/class/struct signatures, container hierarchies (ContainerClass.method_name), clean symbol kind, and ast_symbol_id.
   • Updated MCP search tools (search_code, search_docs) to render rich markdown headers with symbol signatures, source permalinks, and percentage score breakdowns.

3. Search Inspector Redesign & UX Polish:

   • Retained full code chunks for inspection without premature truncation.
   • Added 3-way search mode selector (Hybrid, Semantic, Lexical).
   • Added interactive hybrid split slider with quick presets (Balanced 50/50, Semantic Bias 70/30, Keyword Bias 30/70, Pure Semantic, Pure Lexical).
   • Added indexed repository dropdown datalist populated from /admin/api/repos.
   • Added score breakdown chips showing total combined score alongside dense and sparse component scores.
   • Implemented 1-click Open in Navigator bridge passing repo, file path, and start/end line bounds directly into CodeNavigator with automatic code viewer highlighting.
   • Added 1-click Copy Code snippet button.

4. Testing & Verification:

   • Backend: All 538 tests passing (including Qdrant hybrid modes, search delegation, and FastMCP tool execution).
   • Frontend: All 32 test suites and 320 tests passing.
   • Synchronized REQUIREMENTS.md baseline (1018 automated tests).