ignore: skip loading unreachable ignore files

[jif-oai]

Fixes #2565.

The serial walker currently loads ignore files even when a directory is already skipped or we've reached the max depth. Since we won't visit any children, that work seems unnecessary.

I see that #2566 tried to fix this by adding a new matcher method and threading an is_leaf flag through the ignore internals. This instead reuses the existing add_child_with_entries(path, &[]), so the change stays in walk.rs and the ignore stack still works as before. Let me know if the scoping is not important here

Local benchmarks show:

• 6–9% faster on a real repo.
• ~3× faster on nested ignored directories.
• 22–28× faster on ignore-heavy or sorted searches.
• No change to parallel traversal.

Also adds a regression test for malformed ignore files at the max depth

[BurntSushi]

This makes sense to me! Thank you.

[BurntSushi]

Nice, clever test.

[BurntSushi]

On my checkout of the chromium repository (which is loaded with gitignores), I do see a small but measurable improvement here:

$ hyperfine 'rg-master -j1 --max-depth 6 regex-internals' 'rg -j1 --max-depth 6 regex-internals'
Benchmark 1: rg-master -j1 --max-depth 6 regex-internals
  Time (mean ± σ):     867.7 ms ±   6.9 ms    [User: 377.1 ms, System: 488.8 ms]
  Range (min … max):   860.4 ms … 879.6 ms    10 runs

Benchmark 2: rg -j1 --max-depth 6 regex-internals
  Time (mean ± σ):     842.3 ms ±   3.7 ms    [User: 367.7 ms, System: 472.9 ms]
  Range (min … max):   836.4 ms … 846.8 ms    10 runs

Summary
  rg -j1 --max-depth 6 regex-internals ran
    1.03 ± 0.01 times faster than rg-master -j1 --max-depth 6 regex-internals

Nice find!