Improve block hashes fetching

[Nashtare]

We currently do around 128 RPC queries to get the list of previous block hashes to pass to the prover.
We may improve on this by persisting the previous block hashes in memory. Upon receiving the next block, we'd just need to rotate the vector and replace the item in position 255 with the parent hash.

Fairly minor improvement vs proving cost though.

[vladimir-trifonov]

in prove_blocks.sh we call

cargo r --release --bin leader -- --runtime in-memory jerigon --rpc-url "$3" --block-number $i --proof-output-path $OUT_PROOF_PATH $PREV_PROOF_EXTRA_ARG > $OUT_LOG_PATH 2>&1

for every block i.
Any ideas on how we can persist it in memory?

[vladimir-trifonov]

maybe we should add a way to call it once with from-block to-block params? @muursh

[Nashtare]

Yeah that wasn't really meant to be used with the helper scripts, but more in production environment when the leader would be most likely be running in the background as a daemon.

Although, kinda orthogonal, but it would be really nice if as you suggested we could tweak prove_blocks.sh to have the entire loop being run in one go, as we currently need to load all the circuits every time which is kinda wasteful (especially as in-memory runs with Monolithic mode by default IIRC?)

[BGluth]

So I actually started some work that allows Jerrigon mode to take in a range of blocks to prove (#60). This is also pretty useful for benchmarking and looks pretty easy to do. If someone else wants this work let me know and I an push what I currently have.

[Nashtare]

Superseded by #325