[refactor] Reuse the same stack for all cycles heads in validate_same_iteration

[MichaReiser]

Retrieve the QueryStack once (including zalsa_local) instead of once for each cycle head

[netlify]

✅ Deploy Preview for salsa-rs canceled.

Name	Link
🔨 Latest commit	e629cc4
🔍 Latest deploy log	https://app.netlify.com/sites/salsa-rs/deploys/67f76abde5d90900089fe725

[codspeed-hq]

CodSpeed Performance Report

Merging #791 will improve performances by 5.59%

_{Comparing MichaReiser:cycle-heads-reuse-stack (e629cc4) with master (b165ba7)}

Summary

⚡ 5 improvements
✅ 7 untouched benchmarks

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	amortized[Input]	3.8 µs	3.6 µs	+5.59%
⚡	new[Input]	10.4 µs	9.9 µs	+4.99%
⚡	mutating[10]	13.9 µs	13.3 µs	+4.37%
⚡	mutating[20]	14.1 µs	13.5 µs	+4.31%
⚡	mutating[30]	14.2 µs	13.6 µs	+4.27%

[MichaReiser]

Wow, that's unexpected. Maybe this is because we now avoid claiming the stack if there are no cycle heads?

[carljm]

Nice! That's odd because I thought I had tried this and seen no difference. But I didn't have the short-circuit if empty; maybe that is the difference? It doesn't quite make sense to me, because we should only enter this function for provisional memos, which mean they have cycle heads.

But anyway, the code looks fine, and certainly can't argue with the benchmark results!

[davidbarsky]

Oh wow, this basically removes the (small, worthwhile!) performance penalty that #603 introduced.