Bump to 4.14.1 for crates.io publish

Update Cargo.lock for v4.14.0

Agent benchmarks, Python indexer improvements, bump to 4.13.0

- Add agent-level benchmark suite: 25 tasks across 5 repos measuring
  Claude Code token consumption with vs without wonk (37.4% reduction)
- Update README benchmarks section with agent benchmark results
- Index Python class-level annotated assignments (e.g. pydantic fields)
- Split Python import_from_statement into per-name references
- Return Vec<Reference> from match_import_ref for multi-name imports
- Add .gitignore entry for __pycache__/

Fix CI formatting in mcp.rs, bump to 4.12.1

File filters, pagination, test exclusion, skip-embed, incremental MCP…

… init, bump to 4.12.0

- Add --file/-f filter to search, ref, show, callers, callees, callpath
- Add --reference-file, --callers-file, --callees-file, --destination-file for disambiguation
- Add --page for paginated budget output, --include-tests global flag
- Add --limit to sym, --output files mode to ref
- Make show name optional when --file is given (file-only browsing)
- Add --skip-embed to update command for structural-only index refreshes
- MCP wonk_init now checks wonk_version and uses incremental path when current
- SessionStart hook uses --skip-embed to avoid unnecessary Ollama activity

Qualified paths, cfg macro expansion, subclass refs, auto-regex, bump…

… to 4.11.0

- Parse qualified names (tokio::spawn) in wonk_sym/wonk_show for Rust module paths
- Pre-expand cfg_*! macro wrappers so tree-sitter can parse inner items
- Include subclass/implementor symbols in wonk_ref results via type_edges
- Auto-detect regex patterns in search queries (backslash escapes, char classes)
- Add --file filter to wonk_sym for path-restricted symbol lookup
- Deprioritize test/bench paths in symbol results
- Update specs: add PRD-REF-REQ-004 (output=files), sync architecture tool manifest (22 tools), add v4.10.0 changelog

MCP token efficiency fixes, bump to 4.10.0

- Soften truncation hints to prevent agent Read fallback ("Showing N of M" instead of "Increase budget")
- Clarify wonk_search is keyword/regex only, not natural language
- Move RRF semantic fusion from wonk_search to wonk_ask (single smart search entry point)
- Add output='files' mode to wonk_ref for compact file-list responses
- Deprioritize .d.ts files in wonk_sym results
- Fix shallow mode for TS interfaces: index method_signature/property_signature with scope
- Update wonk_show description to discourage redundant Read calls
- Add wonk_ask to MCP server instructions

Auto-shallow fallback in wonk_show on budget overflow, bump to 4.9.0

When a container type's full body exceeds the token budget, automatically
retry in shallow mode (signature + child signatures) instead of returning
empty results. Adds auto_shallow field to ShowOutput so agents know the
result was downgraded. Hints now suggest shallow:true instead of Read.

Extract doc comments from source, simplify summary detail levels, bum…

…p to 4.8.0

- Add doc_comment field to Symbol: tree-sitter extraction for all 12 languages
  (Rust ///, Go //, Python docstrings, JSDoc /**, etc.), truncated to 200 chars
- Replace Rich/Light/Symbols detail levels with Outline/Rich:
  - Outline (new default): top-level types + functions with doc comments, no methods
  - Rich: all symbols in tree hierarchy including scoped methods
  - Backward compat: "light" and "symbols" parse as Outline
- Auto-generate LLM descriptions (no more --semantic flag): graceful degradation
  when Ollama is unreachable, directory overview prompt from children
- MCP wonk_summary simplified: removed detail/semantic/tree params, always Outline
- Schema migration: doc_comment TEXT column on symbols table

Improve MCP tool descriptions for better agent discovery, bump to 4.7.0

Add instructions field to MCP InitializeResult to guide agents toward
wonk tools over Glob/Read/Grep. Rewrite tool descriptions with explicit
"replaces X" and "more precise than Y" signals to improve discovery and
correct usage (e.g. wonk_summary with depth=1 instead of per-file calls).