perf(root): reveal proofs separately from state updates

[shekhirin]

Unbundle sparse trie proof revealing from state updates, so that we can reveal proofs out-of-order and make the last reveal of the slowest proof as fast as possible. State updates for the sparse trie should still be applied in the order of sequence numbers, because updates can depend on each other.

This PR changes the ProofSequencer structure into the StateUpdateSequencer, and reveals proofs as soon as they come, only sequencing the state updates that should be applied in order.

I don't have a particularly good block on mainnet that will show the improvement, so here's just last 10 mainnet blocks. There's basically no difference, but in theory it should improve on situations when a slow proof is delaying not just all state updates, but also revealing other proofs.

Mainnet Block	newPayload latency @ main	newPayload latency @ this branch
21744021	826.29157ms	813.968945ms
21744022	89.802651ms	72.406547ms
21744023	100.292771ms	89.415594ms
21744024	59.863341ms	45.230231ms
21744025	148.03544ms	151.032709ms
21744026	30.603106ms	33.761105ms
21744027	308.403489ms	320.999399ms
21744028	117.503467ms	128.529333ms
21744029	72.660279ms	66.99271ms
21744030	54.977164ms	51.821672ms

[codspeed-hq]

CodSpeed Performance Report

Merging #14077 will not alter performance

_{Comparing alexey/state-root-task-proof-reveal-separate (1e5ff35) with main (bd5dde5)}

Summary

✅ 77 untouched benchmarks

[shekhirin]

This does not significantly improve the latency, because revealing proofs is already very fast. We can try applying both proofs and state updates out-of-order in another PR.