Try multiprocessing parallelism improvement

[MeGaGiGaGon]

Description

Trying the change from #4771 , I'd like to see if this helps reduce the diff-shades runtime, since the 22% speed increase from #4771 could be enough to make the action consistent again.

Checklist - did you ...

• Implement any code style changes under the --preview style, following the
  stability policy?
• Add an entry in CHANGES.md if necessary?
• Add / update tests if necessary?
• Add new / update outdated documentation?

[JelleZijlstra]

Seems fine to do, could also add a changelog entry since this likely will help perf in some cases.

I think Shantanu commented that we could reuse stat results from the cache logic, that might speed it up a bit more (avoiding syscalls).

[github-actions]

diff-shades reports zero changes comparing this PR (2c67c00) to main (b895a73).

---

What is this? | Workflow run | diff-shades documentation

[cobaltt7]

Looking at diff-shades, there is a noticeable difference in speed when running bigger files first. Stable style went from 6:51 to 4:23. However, preview style only went from 7:19 to 7:10.
Also, it may not be worth it to try reusing the stats from the cache logic; they have to be passed between functions in weird ways. I got it working currently, but the cache tests need to be changed.
I'm not sure how reliable the old diff-shades run from before stats were reused is, but in that run, stable style increased from 6:13 to 6:23.