perf(files): reuse file stats for size filtering and adaptive concurrency

[luojiyin1987]

Summary

• Read file metadata once when size filtering or adaptive concurrency needs it.
• Reuse retained file metadata for adaptive concurrency.
• Keep the existing stat concurrency limit of 128.

The duplicate scan only occurs when both --max-file-size and --threads auto are enabled.

Preserved behavior

• Size filtering runs before adaptive concurrency.
• Filtered large files do not reduce the worker count.
• Fixed threads without a size limit perform no metadata scan.
• Stat failures still propagate to the caller.

Benchmark

Each value is the median of seven warm runs on local ext4.

Files	Before	After
1k	55.7 ms	32.4 ms
10k	560 ms	313 ms
50k	2.82 s	1.57 s

The measured cost scales approximately linearly with the file count.

Validation

• npm run lint
• npm test -- --runInBand
• 26 test suites passed.
• 222 tests passed.
• git diff --check

Security

• This change does not include logs or secrets.
• No redaction was required.