perf(trie): parallel rlp node updates in sparse trie

[shekhirin]

Makes the RevealedSparseTrie::rlp_node take a non-mutable reference to self and return the list of updates that are need to be applied to the trie nodes. This allows us to run multiple rlp_node in parallel in the RevealedSparseTrie::update_rlp_node_level, because we gather the nodes at the provided level and calculate their hashes independently from each other.

[rkrasiuk]

feature gate behind rayon or parallel or smth pls

[codspeed-hq]

CodSpeed Performance Report

Merging #13251 will degrade performances by 11.01%

_{Comparing alexey/sparse-trie-parallel (0f07a91) with main (905fd37)}

Summary

❌ 4 regressions
✅ 73 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
❌	size 100000 | updated 1% | depth 5	10.2 ms	11.4 ms	-10.07%
❌	sparse trie[init size 1000 | update size 100 | num updates 1]	1.3 ms	1.4 ms	-10.97%
❌	sparse trie[init size 1000 | update size 1000 | num updates 1]	5.7 ms	6.4 ms	-11.01%
❌	sparse trie[init size 10000 | update size 1000 | num updates 1]	12.7 ms	14.2 ms	-10.52%

[fgimenez]

with these changes the type returned doesn't match the method name. It is also quite complex, looking at the type itself it is not obvious for instance what the usize is, wdyt?

[shekhirin]

This PR is outdated, will be revised later.