lib/trie: limited parallelism for hashing of nodes

[qdm12]

Task summary

The current parallel hashing implementation hashes each 16 nodes in parallel and recursively.
This can cause memory pressure on the stack since, although cheap, goroutines are not totally free.

The hashing should be changed to have a limited parallelism, ideally set to the number of CPU cores available.
We should also prioritize parallel hashing of branches to have less context switching and longer lasting goroutines.

Progress

• Parallel hashing was disabled in fix(trie): memory leak fix in lib/trie #2009
• Parallel hashing is being improved/fixed in chore(lib/trie): opportunist parallel hashing #2081