Parallel pippenger from blst

[belijzajac]

I found out that there already exists a Pippenger's implementation for supranational/blst rust binding, which appears to use buckets (a large grid for storing the results) and a thread pool for parallelization.

Implementing it achieved the following results:

Benchmark	Old	New	Speedup
blob_to_kzg_commitment	54.145 ms	17.132 ms	3.16x
compute_kzg_proof	65.029 ms	26.509 ms	2.45x
compute_blob_kzg_proof	65.680 ms	27.342 ms	2.40x
bench_commit_to_poly	348.61 ms	97.402 ms	3.58x

However, there seemed to be a little caveat that it didn't work for certain lengths:

• 0 --> panics with message scalars length mismatch
• 1 --> fails the test_fk_multi_chunk_len_1_512 test
• 6 --> fails 1 out of 222 test vectors

So we opted out for a naive approach for lengths less than 8.

Related issue: #202