Lazy finalization of cycle participants in maybe_changed_after

[MichaReiser]

I'm not entirely sure if we should make this change but it does feel closer to how cycles work in execute. On the other hand, deep_verify_memo now works fairly different for cycles than it does for other queries.

This PR removes the loop from deep_verify_memo. We used the loop to iterate a second time if all dependencies in a cycle remain unchanged. I don't think this is strictly necessary because we can rely on Salsa lazily verifying the inner queries when they're pulled as part of another query (they won't run into a cycle again because it's already verified).

We may want to introduce a new event, e..g DidProvisionallyValidateMemoizedValue (open to suggestions) to make it more apparent that the dependencies were verified, they simply were never "finalized"

I still need to update the comment but I'm interested in what people think first.

Performance

The performance on ty is about the same as the old version

[netlify]

✅ Deploy Preview for salsa-rs canceled.

Name	Link
🔨 Latest commit	87d4e19
🔍 Latest deploy log	https://app.netlify.com/sites/salsa-rs/deploys/681da4e9b6471400084e57f7

[codspeed-hq]

CodSpeed Performance Report

Merging #854 will not alter performance

_{Comparing MichaReiser:cycle-no-deep-verify (87d4e19) with master (af69cc1)}

Summary

✅ 12 untouched benchmarks

[carljm]

Does this make a perf difference in ty?

I think this makes sense.

[MichaReiser]

Does this make a perf difference in ty?

There's no change (within the noise margin)