parallel tests and faster runners

[Stebalien]

fixes #498

[Stebalien]

Ok, so... faster. But we don't have a 4x arm64 runner.

[Stebalien]

The Linux x86 tests are down from 40m to 20m and the arm64 tests are down from 40m to 30m. Both should be faster but that requires improved caching (#505 #506).

[BigLep]

Looks good to me?

I assume we're good to no longer have

--test-threads 1 and -p 1

?

Do we know what changed to allow us to not need this anymore (great that we don't). (No need to slow down to explain to me - I was just curious).

[Stebalien]

I assume we're good to no longer have
--test-threads 1 and -p 1

Looking through history, that was added way back in 2019 (possibly before, it could have been ported from rust-fil-proofs-ffi). See 3de4c15. I'm guessing we may have been testing with GPUs at that point and couldn't use the same GPU from multiple threads? I'm not sure, but it doesn't seem necessary now and removing it makes everything faster.