Better thread management

[hai-rise]

• A thread pool (of parked threads) may work better. Early experiments with rayon were very poor, so we may need to roll a specialized implementation.
• We should start post-processing as soon as the scheduler flags completion, instead of waiting for the scoped threads to join (very sad tail-end latency).
• We should re-consider exiting idle threads early like having no new work after 100 iterations.

[byhashdong]

@hai-rise Why rayon has a poor performance, what's the bottleneck?

[hai-rise]

@byhashdong I didn't have time to inspect; just replaced thread::scope with rayon::scope and saw the benchmarks got much slower. I'd love to add a rayon thread pool to the new Pevm struct (didn't exist back then) and retry it. Kinda similar to how Aptos's doing it: https://github.com/aptos-labs/aptos-core/blob/f8eef74f9f712fcc0e809265ec2b77013a683184/aptos-move/block-executor/src/executor.rs#L72

I can create a dedicated issue for the experiment if you're interested in it 🙏.

[byhashdong]

Sure, I would like to study it. Please assign it to me!