chore: drop slotmap and simplify locking

[arendjr]

Summary

This PR drops the slotmap dependency in favor of papaya, and simplifies some of the locking inside the workspace settings.

The main tradeoff this makes is that we need to perform more cloning of the Settings struct, which is a rather large one. To mitigate the impact of this I changed Matcher such that is holds an Arc internally, making it cheap to clone. As part of this, I was also able to get rid of Matcher's internal RwLock by utilizing papaya again.

Test Plan

CI should remain green.

Also checked our release build against the unleash repository again. These are performance numbers I'm seeing on that repo (I re-ran each a couple of times to make sure these numbers aren't heavy outliers):
Biome 1.9.4: Checked 4511 files in 729ms.
next: Checked 4511 files in 744ms.
This PR: Checked 4511 files in 540ms.

Looks like this creates a nice boost :)
Note that both next and this PR now do the scanning upfront, but that doesn't yet result in any speedup of the analysis step of course...

Disclaimer: I don't really know whether the speedup is the result of dropping slotmap specifically, or because I dropped the "caching" of language settings, or maybe the primary speedup is less locking inside Matcher...

[codspeed-hq]

CodSpeed Performance Report

Merging #4696 will not alter performance

_{Comparing arendjr:drop-slotmap (8b6eaf2) with next (f5ba162)}

Summary

✅ 97 untouched benchmarks

[ematipico]

The changes seem to be straight forward! And it seems we got a small boost too, at least we save something before slowing things done with the multi analysis :D

I just left a question