perf(tree): chunk multiproofs

[shekhirin]

Towards #14417

Problem

Given proof targets with 100 accounts, it's preferred to spawn 10 multiproofs with 10 accounts each, instead of spawning one multiproof for all 100 accounts at once. The reason for that is that 10 multiproofs can be executed in parallel (even considering that they may do redundant work because of the intersecting trie nodes), while one multiproof will be executed serially for all accounts.

Same idea can be applied to storage slots: proof targets with 1 account and 100 storage slots is better spawned as 10 multiproofs for 10 storage slots each, rather than one multiproof for all 100 storage slots at once.

Solution

This PR adds chunking on the MultiProofTask side.

1. In on_prefetch_proof, proof targets are chunked and spawned as separate multiproofs.
2. In on_state_update, state update is chunked, proof targets are created for each chunk, and spawned as separate multiproofs.

Benchmark

[codspeed-hq]

CodSpeed Performance Report

Merging #14800 will not alter performance

_{Comparing alexey/chunk-proof-targets (f9ecff0) with main (53e72c8)}

Summary

✅ 77 untouched benchmarks

[mattsse]

I can follow the logic here, defer to @rkrasiuk re chunking

looks sound to me