feat: Implement #599

[Veykril]

Currently looking into #599, this change makes things a bit simpler to overview removing some back and forth in code paths

[netlify]

✅ Deploy Preview for salsa-rs canceled.

Name	Link
🔨 Latest commit	5c7a7c5
🔍 Latest deploy log	https://app.netlify.com/sites/salsa-rs/deploys/67cadb52f1cbd6000868832c

[codspeed-hq]

CodSpeed Performance Report

Merging #748 will improve performances by 3.28%

_{Comparing Veykril:veykril/push-zmlovtwvwxty (d1f6e52) with master (9d2a978)}

Summary

⚡ 1 improvements
✅ 10 untouched benchmarks

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	amortized[Input]	3.7 µs	3.6 µs	+3.28%

[Veykril]

So implementing the memo table part is actually trivial, the real issue is the dependency tracking as that goes through the type erased ingredient trait and now we have potentially more keys than just an ID ...

[Veykril]

I just realized, all of this is probably unnecessary! Instead of interning the additional arguments, can't we just synthesize a new tracked struct with each arg being a field? That would trivial fix things from what I can tell?

[MichaReiser]

That won't work because you can't create tracked structs from outside a query. Each tracked struct also gets a new id (it's uniqueness is defined by its instance not only its values). It would limit multi argument queries to only be callable from other queries.

It's also not clear to me how that's different from interning?