Benches for wave spawning

[ElliottjPierce]

Objective

Establish benchmarks for waves of entity spawning as discussed in #18054. This does not include (for now) wave insertions, as mere insertions don't affect entity allocation.

Solution

Add benchmarks for permutations of entity count, wave count, component size, and batch spawning function.

The benchmarks do not show much difference in the component size variance, so I would be happy to remove that and decrease complexity if desired.

Showcase

Yikes! Here are the results on M2 MAX:

(Ignore the regressions and improvements. That's from testing the benches.)

Here, I only ran the EmptyComponent benches.

With only one wave, spawn_batch and insert_or_spawn_batch both run in roughly 3ms because there are no shenanigans in Entities regarding allocation.

So if one wave takes 3ms, 2 waves should take 6ms, right? Nope! spawn_batch on 2 waves takes 6ms, but insert_or_spawn_batch with the infamous alloc_at_without_replacement takes 760ms!!. That's a little more than 6.

Running 16 waves was taking too long, but I'll leave it running for a bit so we can get a number. I would estimate ~11 seconds (700ms per wave on average) for insert_or_spawn_batch. This implies that the relationship isn't quite linear since the above 2 waves was also 700ms. But this could be memory allocator shenanigans too.

EDIT: It did end up being ~11.5 seconds.

[alice-i-cecile]

Nice stuff! Now that we've isolated the problem in question to an API that we want to cut anyways, I'm a bit iffy on adding these benchmarks. This specific problem feels very bizarrely idiosyncratic. While this PR was 100% worth the time to demonstrate the problem, I'm not sure that this set of benches is going to catch any new problems.

The benchmarks do not show much difference in the component size variance, so I would be happy to remove that and decrease complexity if desired.

If we decide to keep this let's cut that.

[ElliottjPierce]

I'm a bit iffy on adding these benchmarks.

Totally agree. If the solution is to remove the think we're benchmarking, what's really the point? Still glad I got the numbers though.

If you do choose to keep this, just ping me, and I'll clean it up. Otherwise feel free to close the pr.

Also can't help but adding that the most recent label change makes I think 100% of my prs and issues contentious or controversial. Rightfully so, but it is funny.