Generate crate_universe bzlmod repositories in parallel

[UebelAndre]

This change aims to improve crate_universe for bzlmod by parallelizing generation into a single module_ctx.execute call to the cargo-bazel binary for each module. This avoids the limitation of running each module serially.

[UebelAndre]

It's interesting that this is roughly as slow as the serial implementation. I suspect the bottleneck is the cargo tree calls which I believe puts a lock on the local registry being used.

Before

After

[UebelAndre]

It's interesting that this is roughly as slow as the serial implementation. I suspect the bottleneck is the cargo tree calls which I believe puts a lock on the local registry being used.

I should point out that this was for a run without REPIN=1. I also misspoke about cargo tree calls, unless splicing is happening (which occurs with REPIN=1), cargo tree should not be called. When all crates are repinning we do see a larger time savings but not as much as I'd expected:

Before

After