[core/state] Prefetch Trie Nodes Concurrently

[patrick-ogrady]

Why this should be merged

Context: ethereum/go-ethereum#28266

How this works

This change modifies core/state/trie_prefetcher.go to prefetch trie nodes in parallel. In addition to this tweak, prefetching is no longer aborted (wait until all keys on a trie are prefetched) and we now prefetch tries during block building.

How this was tested

• Existing unit tests
• Run on the Avalanche Network

Open Questions

• What should the default max prefetch concurrency be? (32 seems low given the size of this data, see example below): [core/state] Prefetch Tries Concurrently (Aggressive=48) #375 -> start with 16
• Should we scale up copies faster? -> use the same scale up approach as bounded workers
• Why are there any skipped prefetch tasks?: 947cad8
• How many trie copies are created per block -> add metric
• Should we reserve a trie just for copying? -> a fetch on an unpopulated path could take 7ms (7 levels), and likely one copy per 7ms
• Should we reduce tasks per copy? -> could take 50ms per group (if broke into sections of 3, would only take 16-20ms), also add a metric for this
• does a lock get grabbed when multiple threads fetch the same node? -> no
• update the base trie to one with the most populated keys from prefetching (approx): [core/state] Return Best Trie Copy #396
• Limit copies per subfetcher (could ensure no single trie is copied too many times, while still being less than disk concurrency)?
• Nits: [core/state] Prefetch Trie Nits #398