perf(trie): use local ThreadPool in Parallel::multiproof

[fgimenez]

Aiming to prevent deadlocks when enabling ParallelProof::multiproof ref #13260 it allows us to process proofs in chunks, so that we don't spin up tasks for all the existing proofs at once, potentially causing resource exhaustion and rayon contention.

the goal of this change is just to prevent the deadlock we were experimenting with when enabling parallel multiproof, but I also wanted to verify that splitting the workload in chunks doesn't add a significant overhead, fixing the time calculation for the state root task we have consistently better times compared to the regular calculation (state root task time is usually a bit above half of the parallel calculation time)

Dec 17 07:39:37 reth-5 reth-holesky[344736]: 2024-12-17T07:39:37.597759Z  INFO engine::tree: State root task finished block=NumHash { number: 2953384, hash: 0xc80f1ab43ca7cbc67c22edd9e32ae6b039efe06dfccc588e1160906259ff616d } total_time=102.27362ms task_state_root=0x1b50b96acf75949b0ace02fcad63ad56242c94cc93ef51f64d8c87309359d330
Dec 17 07:39:37 reth-5 reth-holesky[344736]: 2024-12-17T07:39:37.664179Z DEBUG engine::tree: Calculated state root root_elapsed=191.658237ms block=NumHash { number: 2953384, hash: 0xc80f1ab43ca7cbc67c22edd9e32ae6b039efe06dfccc588e1160906259ff616d }

[shekhirin]

nits, @rkrasiuk PTAL