perf: Collect module symbols in parallel

[ChayimFriedman2]

This saves 2 seconds when cache priming omicron on my computer.

[Veykril]

That suprises me to be honest as that sounds like we aren't utilizing a lot of threads in cache priming?

[ChayimFriedman2]

First, I have a lot of cores in my machine (32 to be precise), so that's maybe a factor (but for people with less cores this probably won't be worse; at most, not better). But also, we populate symbols for local crates, and at that point, the dependency graph narrows and we indeed don't have a lot of parallelism.

[Veykril]

Well my thinking is that the module symbols query should not depend on other module symbols, and at this stage we already have populated all def maps. So the graph narrowing should not matter that much. Not opposed to this here though, just something I am a bit confused by.

[ChayimFriedman2]

I realized we don't put cache priming in rayon threads, so this may cause conflicts. I will fix this.

[Veykril]

Closing in favor of #19721