feat: consume new ssz batch hash branch

[twoeths]

Motivation

• reduce hash_tree_root time in lodestar, especially at epoch transition time
• this leverage the new hashtree which and new as-sha256 which support batch hash api

Description

• consume te/batch_hash_tree_root ssz branch, the critical update there is using hashtree and minimal memory allocation to compute tree root
• ssz PR: feat: batch hash tree root ssz#378
• this follows the approach 1 as noted here
• copy the implementation of ListValidatorTreeViewDU from the ssz branch, the main thing is to compute validators' roots in batch during commit() phase
• hashTreeRoot() is improved in 2 ways:
  • batch hash validators, execute HashComputation[] in batch
  • memory optimization, a lot of metrics are better (see below)

part of ChainSafe/ssz#355
closes #6598

TODOs

• Investigate why do hashTreeRoot() inside state transition causes more memory and gc? right now we do the hashTreeRoot() in beacon-node instead. Will retry when we have a published version because right now I only consume local ssz branch on test nodes. Update: resolved in fix: batch hash tree root in state transition #7032
• after a BeaconState hashTreeRoot() is computed, if we call some read-only operations, like state.balances.length or state.validators.length, it will rebind and recompute the tree root again. Need to find a way to avoid that, see ViewDU container rebinds node with read-only operations ssz#379

[twoeths]

posting some great metrics (the test mainnet node) captured from feat1 - 6h rate interval

• Prepare Next Epoch time is ~600ms faster than beta/unstable and ~500ms faster on mainnet

• Beacon block gossip validation time is 50ms faster on mainnet node and 60ms faster on holesky

• Execution api notifyNewPayload is 30ms - 40ms faster on mainnet and 60ms faster on holesky

• Block transition is ~5ms faster on both holesky and mainnet

• This also reflected in the Avg block state transition time in verifyBlocksStateTransitionOnly.ts: 35ms faster on mainnet and 50ms faster on holesky

[twoeths]

• "till become head" is ~30ms faster than stable on mainnet, somehow it's the same to beta mainnet node

• it's ~90ms faster on holesky (both beta + mainnet)

• "avg block recv to import delay" is consistent, it's 50ms - 60ms better than both beta and mainnet node

these metrics are not the accumulation of all above better metrics because they also depend on blob I guess

[twoeths]

a side effect from this branch is "Process block commit step avg time" is usually higher than stable (<= 5ms) because we also compute validators' root when commit. This metric and "Process block avg time" are not as important as "Avg block recv to state transition delay" (which is >=50ms better than stable/unstable), because we have to compute state root anyway

[twoeths]

closing in favor of #7171