Improve lagrange retrieval performance & remove unsafes

[volhovm]

1. This creates a binding function _lagrange_commitments_whole_domain that is to be used instead of calling _lagrange_commitment (that retrieves a single base) domain_size times.
   • The optimisation saves up about 100ms on every prove call on my local machine if I remember correctly, or less -- in the end it cuts about 0.5m or 6% time on the cluster:
   • Benchmark improvement (on a fast-ish cluster) for _build/default/src/app/cli/src/mina.exe transaction-snark-profiler --zkapps --k 1 --max-num-updates 4 --min-num-updates 2:

- NEW
  - 6.231m
  - 6.378m
  - 6.381m
- OLD:
  - 6.843m
  - 6.753m
  - 6.743m

2. Additionally I removed some unsafes that are not anymore necessary since SRS is now wrapped in a Mutex (which should be RWLock really).

Caveat: this change plus some amount of debug printing invokes a completely different deadlocking bug with me where the verifier process no longer starts. This seems to be triggered by the code becoming faster, so some race condition is triggered.

I managed to reproduce the deadlock over compatible too: o1-labs/proof-systems#2728

[volhovm]

!ci-build-me

[volhovm]

!ci-build-me

[volhovm]

!ci-build-me

[volhovm]

!ci-build-me

[querolita]

Caveat: this change plus some amount of debug printing invokes a completely different deadlocking bug with me where the verifier process no longer starts. This seems to be triggered by the code becoming faster, so some race condition is triggered.

Do you mean we are paying ~100ms less but now we have a new race condition because a thread becomes too fast and accesses a resource? Then I am not sure we should merge this before we solve the other problem first 🤔

[volhovm]

@querolita

1. I ran tests on this branch many times and they all succeed. So I don't think there's anything in this thread that is actually a problem.
2. By adding some printing I managed to reproduce the deadlock even over compatible. This makes me suspicious that the problem is with printing on the rust side that breaks some pipes with the verifier client, and it's not lagrange related at all.
   • See Deadlock while creating verifier process o1-labs/proof-systems#2728
3. I'm happy to take responsibility for any issues if they come up

[querolita]

I find it a bit confusing that the one using Pasta_bindings.Fq.t is called caml_fp_srs_lagrange... , but I see this is the format being followed across this file so I suppose that notation is expected.

[volhovm]

!ci-run-nightly

[volhovm]

!ci-nightly-me

[volhovm]

I will check if nightlies pass first, and merge if they do. I'm fairly confident it's OK but I'm also not in a hurry.

[volhovm]

Nightlies seem to pass (I did it twice) except for "Publish Mina for Bullseye" which fails due to some unrelated debian thing.

Merging.