feat(config): map custom file extensions to languages via codegraph.json (#906) (#955)

The extension → language table was hardcoded, so a codebase using a
non-standard extension for a supported language (e.g. `.dota_lua` for Lua)
had those files silently skipped — no way to opt them in short of patching
the source.

Add an opt-in, project-scoped `codegraph.json` at the repo root:

    { "extensions": { ".dota_lua": "lua", ".tpl": "php" } }

Mappings merge on top of the built-in defaults and take precedence (so a
built-in can be re-pointed, e.g. `.h` → `cpp`). Absent or malformed config
is the zero-config default — byte-identical to prior behavior; an invalid
target language or unparseable file is warned-and-skipped, never fatal.

Implementation:
- New `src/project-config.ts` — `loadExtensionOverrides(rootDir)`, validated
  against `isLanguageSupported`, mtime-cached per root.
- `detectLanguage` / `isSourceFile` gain an optional `overrides` arg
  (omitting it is the existing behavior).
- Overrides threaded per-operation through every extraction call site
  (scan/walk gates, git change-detection, grammar selection, extraction,
  the file watcher), resolved from the project root — no process-global
  state, so the multi-project daemon stays isolated. The parse worker
  receives the resolved language in its message.

Tests: 13 new cases (unit, loader validation/normalization/caching, and a
full-index integration proving a custom-extension file is extracted while
the zero-config path indexes nothing). Worker path smoke-tested via the
built CLI.

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

src/extraction/parse-worker.ts

@@ -55,14 +55,17 @@ import type { Language, ExtractionResult } from '../types';
 const PARSER_RESET_INTERVAL = 5000;
 const parseCounts = new Map<Language, number>();
 
-parentPort!.on('message', async (msg: { type: string; id?: number; filePath?: string; content?: string; languages?: Language[]; frameworkNames?: string[] }) => {
+parentPort!.on('message', async (msg: { type: string; id?: number; filePath?: string; content?: string; languages?: Language[]; frameworkNames?: string[]; language?: Language }) => {
   if (msg.type === 'load-grammars') {
     await loadGrammarsForLanguages(msg.languages!);
     parentPort!.postMessage({ type: 'grammars-loaded' });
   } else if (msg.type === 'parse') {
     const { id, filePath, content, frameworkNames } = msg;
     try {
-      const language = detectLanguage(filePath!, content);
+      // The main thread resolves the language (it holds the project's
+      // codegraph.json extension overrides) and sends it; fall back to detection
+      // for older callers / safety.
+      const language = msg.language ?? detectLanguage(filePath!, content);
       const result: ExtractionResult = extractFromSource(filePath!, content!, language, frameworkNames);
 
       // Periodic parser reset to reclaim WASM heap memory