fix(scanner): bound scans by file count; refuse the dependency scan above it and say so

[JordanCoin]

Measurements (macOS, /usr/bin/time -l, synthetic TypeScript repos, one import per file, default bound 100,000)

repo	command	before: wall	before: max RSS	after: wall	after: max RSS
60k files	codemap .	1.42 s	53 MB	0.37 s	51 MB
60k files	codemap --deps	5.34 s	421 MB	5.02 s	421 MB (under the bound, runs as before)
60k files	codemap --importers f0.ts	7.26 s	422 MB	4.33 s	424 MB
200k files	codemap .	4.25 s	127 MB	4.13 s	70 MB (keeps 100k, counts 200k)
200k files	codemap --deps	29.57 s	1,565 MB	3.07 s	113 MB (refused)
200k files	codemap --importers f0.ts	14.52 s	1,634 MB	1.74 s	113 MB (refused)

Memory grows linearly with file count, about 8 MB per 1,000 files on the dependency path, which is how two concurrent scans of several large repos exhausted the machine in the report. Before this PR nothing refused, warned, or degraded. The default is 100,000 files: on these numbers that is roughly 800 MB and 10 s, the point where a default should stop and ask; 50,000 would have turned away repos a laptop handles in 5 s.

What changes

• Walk bound. ScanFilesBounded keeps at most limits.DefaultMaxScanFiles (100,000) files and keeps counting, so the tree prints the whole count. ScanFiles is unchanged for callers.
• Dependency scan refused above the bound. ast-grep walks the tree itself and its full JSON is held in memory, so the only place to bound it is before it starts. scanForGraphOutcomeWithFilters, which every graph build routes through (ScanForDeps, BuildFileGraph), counts first and above the bound records a scan-bound source (status unavailable) and runs nothing. --importers, blast-radius, find, hooks and MCP read the same coverage.
• Honest wording. --importers says Importers of X: unknown, no dependency scan ran. rather than No files import X.; --deps says Dependency scan not run: repository is above the file bound. rather than No source files found.; MCP get_importers gets the same unknown wording.
• Raise it. --max-files N per run, max_files in .codemap/config.json per repo; -1 removes the bound. README documents both.

Exact coverage lines on the 200k repo:

Coverage: partial — bounded: scanned 100000 of 200001 files (limit 100000; raise with --max-files or config max_files)
Coverage: unavailable — bounded: scanned 0 of 200001 files (limit 100000; raise with --max-files or config max_files)

JSON: Project.scan_bound {scanned, total, limit} (omitted when not bounded).

Tests

• TestScanFilesBoundedStopsAtLimit, TestScanForDepsRefusesAboveBound (scanner/scan_bound_test.go): explicit bounds of 3 and 2 over tiny fixtures, never the default; bound of 2 over 3 TS files produces no analyses, coverage unavailable with the exact note, the graph has no edges; at the bound nothing changes.
• TestImportersReportUnavailableSaysUnknown (scan_bound_cli_test.go): no "No files import" when nothing was scanned.
• go build ./... && go test ./... && gofmt -l . clean.

Not changed

No edge cap (edges are bounded once files are), no runtime.MemStats sampler (the bound prevents the allocation; a sampler would only report after the fact), no wall-clock scan timeout (the existing ast-grep timeout and hook budget remain), no refusal to start the watch daemon (it inherits the bounded file list through ScanConfiguredFilesWithFilters).

Fixes #186

🤖 Generated with Claude Code

https://claude.ai/code/session_01NRaBgKRAZMX9hjbkkefA7T

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.