Improve performance of Cache::analyze

[aumetra]

Improve the performance of the Cache::analyze function.
When analyzing the current benchmarks using perf, you'll see that most of the time is spent inside the FuncValidator::valdiate function.

This PR changes the logic to only validate the function bodies when the WASM is saved into the cache since we already assume that the cache only contains modules which check all the validation boxes.

Now the largest timespan spent is on the SHA-256 checksum generation.

Another minor change is that the function body validation is now done in parallel which will mostly benefit WASM modules with a lot of functions.
The downside of this is that the allocations can't be reused since that would imply shared mutable ownership across threads, and using a mutex here would just make this code essentially sequential again.

But there is still an ~15% performance gain over not doing it in parallel on average in my non-reliable benchmarks (they were performed on my workstation with rust-analyzer, Firefox, etc. open).

Closes #2033

[webmaster128]

🙌

[webmaster128]

Now the largest timespan spent is on the SHA-256 checksum generation.

That's surprising to me. Loading the file from disk is much faster than hashing it? Strange. But we probably talk about very small values at this point anyways.

[aumetra]

I am not quite sure how this change could even have impacted the cosmwasm_check unittests? They seem unrelated on a first glance?

[webmaster128]

I am not quite sure how this change could even have impacted the cosmwasm_check unittests? They seem unrelated on a first glance?

I think what is happening here is that the corrupted.wasm now returns a different error because the order of checks in check_wasm changed. I think we can just update the test.

[webmaster128]

Very nice. Could you add a CHANGELOG.md entry to the Unreleased section, that briefly explains what we did here and links the PR?

[aumetra]

Okay, pushed on wrong branch. One sec.