refactor: Move verified_final from Memo into QueryRevisions

[Veykril]

This saves an entire usize per function Memo

[netlify]

✅ Deploy Preview for salsa-rs canceled.

Name	Link
🔨 Latest commit	9eef15b
🔍 Latest deploy log	https://app.netlify.com/sites/salsa-rs/deploys/67ddb02b59316700086fe32b

[codspeed-hq]

CodSpeed Performance Report

Merging #769 will improve performances by 3.35%

_{Comparing Veykril:veykril/push-xoyllxvompzq (9eef15b) with master (6cfe2aa)}

Summary

⚡ 1 improvements
✅ 11 untouched benchmarks

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	accumulator	4.5 ms	4.3 ms	+3.35%

[carljm]

Nice perf improvement! I considered trying this but thought it better to keep QueryRevisions immutable; but now that I look at this PR, I don't really see any strong reason for that.

Why does it save memory? Is it just a matter of the padding/alignment?

[Veykril]

Why does it save memory? Is it just a matter of the padding/alignment?

Yes QueryRevisions has some free padding this can fit in, where as Memo does not usually (it depends on the V type in it).

[Veykril]

Rebased and now the numbers are randomly red in report_tracked_read again 😩 I fear we might be within limits of where cache misses become very obvious in the benchmarks, given report_tracked_read goes through two heap allocs (querystack -> edge indexset).

Thus I'd personally say we should discard the regression here, I don't see it as meaningful. (and we do have to keep an eye out for memory right now, as r-a has had some bad regressions after all)