perf(trie): use rayon for parallel storage proofs

[shekhirin]

This removes the mpsc channels from the ParallelProof::multiproof implementation to solve issues with rayon threads getting blocked on rx.recv() when ParallelProof::multiproof is spawned inside a rayon task.

Instead, we calculate storage proofs in parallel using rayon iterators, collect all results, and then query them when calculating the account trie leaf nodes.

[shekhirin]

The difference with the previous impl is that we now first wait until all storage proofs are calculated, and only then query them when building the account leaf nodes. In the previous impl, we started all storage proof calculations at once, and then queried in the same order, waiting on the channel if they're not ready yet.

I think we should be able to transform it further using rayon iterators, so that we do not block until all storage proofs are calculated.

[gakonst]

Yeah seems like we don't want to block on the parallel storage root and ideally you can background spawn them? This should be diff from the previous implementation given we are no longer traversing left to right and we can do updates unsorted?

[shekhirin]

Me and @fgimenez were running nodes with #13260 for a while now, and it doesn't seem to have any problems. Closing for now.