feat(extraction): make MAX_FILE_SIZE configurable via CODEGRAPH_MAX_FILE_SIZE (#1016)

[maxmilian]

Closes #1016.

What

The 1 MB file-size skip threshold in src/extraction/index.ts was a hardcoded const, so repos with legitimately large hand-written sources (1–5 MB) had no way to get them indexed.

This adds a resolveMaxFileSize() helper that honours a CODEGRAPH_MAX_FILE_SIZE environment variable (in bytes), mirroring the existing resolve*() env idioms in src/mcp/daemon.ts:

export function resolveMaxFileSize(): number {
  const raw = process.env.CODEGRAPH_MAX_FILE_SIZE;
  if (raw === undefined || raw === '') return DEFAULT_MAX_FILE_SIZE;
  const parsed = Number(raw);
  if (!Number.isFinite(parsed) || parsed <= 0) return DEFAULT_MAX_FILE_SIZE;
  return Math.floor(parsed);
}

Unset / non-numeric / non-positive values fall back to the 1 MB default, so existing behaviour is unchanged unless the var is explicitly set. Both existing size-check sites (single-file and bulk paths) are unchanged — they read the same module MAX_FILE_SIZE.

Why this is the whole change

The issue also suggested complementary WASM memory-safety (recycle the worker after large parses). That guard already exists — recycleWorker() is already invoked after large-file parses — so making the threshold configurable is safe on its own and needs no lifecycle changes.

Tests

• New __tests__/resolve-max-file-size.test.ts — default fallback, valid override, fractional flooring, and invalid/zero/negative/non-numeric fallback (9 cases, all pass).
• __tests__/extraction.test.ts — 377 pass, no regression.
• tsc --noEmit — clean.

README updated with the override next to the existing 1 MB documentation.