feat(cli): add --max-file-size flag to index/sync/init (#369)

[eddieran]

Summary

Closes #369.

CI pipelines that only want source-code symbols (and would rather skip multi-MB media/vendor blobs) currently have to patch the codebase: the 1 MiB threshold lives as a compile-time constant in src/extraction/index.ts. This PR makes that threshold configurable per invocation, with the same default so existing workflows are unaffected.

New surface

CLI — three commands grow an identical --max-file-size <size> flag:

codegraph init -i --max-file-size 500kb
codegraph index --max-file-size 2mb
codegraph sync --max-file-size 1.5gb

<size> accepts:

Input	Bytes
1048576	1 048 576 (raw bytes)
500kb / 500 KB	512 000
2mb / 2 MB	2 097 152
1.5gb	1 610 612 736
700KiB / 1MiB	IEC suffixes also supported

Both decimal and IEC suffixes resolve to the binary base (×1024). That matches what du -h / ls -lh report and keeps the 1 MiB default the codebase has always used. Invalid values fail fast:

$ codegraph index --max-file-size banana
✗ Invalid --max-file-size value: "banana". Use a non-negative number with an optional unit suffix (e.g. "500kb", "2mb", "1.5gb", "1048576").

Library — IndexOptions.maxFileSize?: number is added to both CodeGraph.indexAll(opts) and CodeGraph.sync(opts). Omitting it keeps the existing 1 MiB default. DEFAULT_MAX_FILE_SIZE (the renamed MAX_FILE_SIZE) is also exported from src/extraction/index.ts for consumers that want to clamp or report against it.

Implementation notes

• The orchestrator's indexAll, sync, indexFile, and indexFileWithContent methods now take an optional maxFileSize argument with the default constant as fallback — additive, non-breaking on the existing positional-arg shape.
• parseFileSize() lives in src/utils.ts (it's small and pure, and may be useful elsewhere in the future). Returns null on bad input so the CLI can format a clean error and exit 1 — silently coercing to the default would have been worse than failing fast for a CI use case.
• No persistence: each invocation provides its own value, the watcher/auto-sync keeps using the default. This matched the issue's "CI pipelines indexing only source code" framing more naturally than a per-project config setting; happy to layer in a config-file knob in a follow-up if there's demand.

Test plan

New __tests__/max-file-size.test.ts (8 tests) covers:

• parseFileSize — plain bytes, kb/mb/gb (with and without spaces, case-insensitive), IEC kib/mib/gib, and explicit null for malformed inputs (empty, negative, unknown suffix, repeated dots).
• CodeGraph.indexAll({ maxFileSize }) — tightened limit drops a ~5 KiB file with code: 'size_exceeded' while keeping the small one; default keeps both.
• CLI flag end-to-end — invalid size string → exit 1 + clear stderr message; --max-file-size 1kb --quiet against the built binary drops the larger file from the index while keeping the smaller one.

Full suite green:

Test Files  50 passed (50)
     Tests  1040 passed | 2 skipped (1042)

(+8 new tests over main; no regressions.)