Move salsa event system into Zalsa

[Veykril]

The existence of Database::salsa_event is incredibly annoying as a lot of internal code takes a &dyn Database solely to do event reporting where as usually a &Zalsa handle would otherwise suffice. This becomes even more annoying when mutable handles are involved as it requires some odd code structuring to avoid mutability xor shared errors, it also makes optimizations more difficult for the compiler due to dynamic dispatches it can't look through.

So this moves the trait method away to a separate callback that can be registered, and if not registered, can be skipped entirely.

[netlify]

✅ Deploy Preview for salsa-rs canceled.

Name	Link
🔨 Latest commit	14ae8c6
🔍 Latest deploy log	https://app.netlify.com/sites/salsa-rs/deploys/681b42160dbdf30008d20e93

[codspeed-hq]

CodSpeed Performance Report

Merging #849 will improve performances by 6.05%

_{Comparing Veykril:veykril/push-nkkzvzyyqwrz (14ae8c6) with master (f78a641)}

Summary

⚡ 2 improvements
✅ 10 untouched benchmarks

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	amortized[Input]	3.3 µs	3.1 µs	+6.05%
⚡	many_tracked_structs	36.8 µs	34.7 µs	+6.03%

[MichaReiser]

I didn't do a line by line review but this looks great.

[MichaReiser]

Huh, I'm pretty sure I wrote one more comment but it seems it didn't make it.

Roughly, it would be nice if Storage::new could remain argument free and we would isntead have a with_event_logger mutator method. I think the challenge with this is that zalsa is wrapped in an Arc. Would it be possible to use our normal cancellation to get a reference to storage and swap the evnet handler?

[Veykril]

Ye that should be doable, though do note the new constructor is new and the previous Default impl still exists