fix: invalidate Cursor cache after missing-createdAt fix

[lincolnwan]

Summary

• Bump Cursor result cache version from 3 to 4 so stale v3 caches are invalidated.
• Respect CODEBURN_CACHE_DIR for Cursor result cache, matching docs and enabling isolated cache tests.
• Add a regression test for old v3 Cursor caches created before the missing-createdAt fix.

Why

This follows up on fe2e622 ("Skip Cursor bubble rows that lack a createdAt timestamp") from PR #321; it invalidates caches generated before that parser fix.

That parser fix made fresh parsing skip Cursor bubble rows without createdAt, but existing ~/.cache/codeburn/cursor-results.json files could still contain v3 cached calls timestamped with parse time.

If the Cursor DB mtime/size did not change, CodeBurn would continue trusting that stale cache and still report phantom current-day usage.

Testing

• npm test -- tests/cursor-cache.test.ts tests/providers/cursor.test.ts tests/providers/cursor-bubble-dedup.test.ts tests/providers/cursor-workspace-breakdown.test.ts
• npx tsc --noEmit

Notes

This does not change Cursor parsing logic. It only invalidates previously poisoned Cursor result caches and adds coverage for that behavior.

[ozymandiashh]

Validation run on macOS arm64. No local project names, paths, prompts, session IDs, costs, or usage details are included here.

Proof:

• npx vitest run tests/cursor-cache.test.ts tests/providers/cursor.test.ts — 11/11 tests passed.
• npm run build — passed.
• git diff --check — passed.
• Claude Opus 4.7 effort max review — PASS.
• Gemini 3.1 Pro Preview review — PASS.

Read-only review notes:

• The cache version bump from 3 to 4 invalidates stale Cursor result caches written before the missing-createdAt fix.
• CODEBURN_CACHE_DIR is scoped to the cache helper and preserves default behavior when unset.
• The new test isolates the cache path in a temp directory and restores the env var after each test.