execution/commitment: reuse branch buffers instead of cloning every read

[awskii]

Commitment allocated a fresh copy on every branch read. TrieContext.Branch cloned each result, and getDeferredUpdate / mergeDeferredUpdate cloned prev and the merged encoded.

Changes

• TrieContext.Branch copies into a per-context buffer; LatestCommitmentState clones at its own call site.
• getDeferredUpdate takes prev through the existing reuseBytes instead of cloning — the nil-vs-empty contract it cited is what reuseBytes already implements.
• mergeDeferredUpdate writes the merged value into a dedicated encodedBuf, so encoded either aliases raw or points at a buffer that survives pooling.
• Tests pin the new contract: the buffer is reused across reads, and nil stays distinct from present-but-empty once it is warm.

Notes

Mainnet archive node syncing from genesis, 16 cores / 125 GB, --experimental.parallel-commitment, settled windows either side of the change:

metric	before	after
end-to-end	55.8 blk/s	75.9 blk/s
bytes.Clone	75.2 MB/s	13.4 MB/s
allocation per block	7.99 MB	6.26 MB
GC per block	6.25 mcore·s	3.21 mcore·s
total CPU	3.12 cores	3.18 cores

CPU is flat: the gain is GC pressure removed from the serial commitment path, not cycles saved.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.