fix(proofs): reduce CompressedSNARK::prove time

[Autoparallel]

Science

The science has told me that the things that contribute to this dominantly are:

• The total circuits we have used in our ROM. We are currently in WIP to reduce circuits further with feat: hash-based JSON circuit web-prover-circuits#41
• The commitment scheme. It appears HyperKZG is far more efficient at this stage. I believe we knew this before, but switched back to Pedersen for proof size worries. From my investigation, these seem to be far less of an issue now.