cleanup: don't do stale check when passing --prune.

[MikeMcQuaid]

This speeds up --prune fairly considerably.

Before:

$ hyperfine --warmup 3 'brew cleanup --prune=365'
Benchmark #1: brew cleanup --prune=365
  Time (mean ± σ):     33.352 s ±  0.654 s    [User: 21.684 s, System: 6.458 s]
  Range (min … max):   31.901 s … 34.096 s    10 runs

After:

$ hyperfine --warmup 3 'brew cleanup --prune=365'
Benchmark #1: brew cleanup --prune=365
  Time (mean ± σ):      6.821 s ±  0.186 s    [User: 2.225 s, System: 3.926 s]
  Range (min … max):    6.578 s …  7.178 s    10 runs

Closes #10177
Fixes #10172

[BrewTestBot]

Review period will end on 2021-01-01 at 14:00:34 UTC.

[BrewTestBot]

Review period skipped due to critical label.

[fxcoudert]

With that merged, the timings for Intel Big Sur runners have gone from:

• cleanup-before 2m50s
• cleanup-after 3m0s

to

• cleanup-before 45s
• cleanup-after 1m20s

I'd have to check what is still taking so long. Even with thousands of files, getting mtime and ctime should not take an entire minute. For reference, even on the ARM runner that has the most cached files, it's taking 5 seconds in the worst case.

[MikeMcQuaid]

I'd have to check what is still taking so long. Even with thousands of files, getting mtime and ctime should not take an entire minute. For reference, even on the ARM runner that has the most cached files, it's taking 5 seconds in the worst case.

I don't think the ARM runner is a good comparison for various reasons:

• it's a new architecture that's been custom-designed to run macOS
• it's not running in a VM
• it's using SSD storage

find -delete could be an acceptable replacement to put in brew test-bot itself instead of brew cleanup --prune.