perf(state-transition): avoid per-validator pubkey lookups in epoch updated

[fridrik01]

I noticed that on our devnet-core after running load testing for a few hours that block times became unstable after several hours. Looking at the logs it seems that we hit timeouts in process_proposal always at epoch boundaries.

When running our load testing harness, it injects thousands of deposits, and each creates a validator entry. Then at every epoch boundary processEpoch iterates the whole registry, and processEffectiveBalanceUpdates did a per-validator pubkey->index lookup plus a single-key balance read. That is ~2N small reads against disk-backed IAVL on Longhorn. With a bloated registry that makes many random I/O reads, and having Longhorn's high latency spikes we are often exceeding the consensus timeouts at epoch boundaries.

To address this, this PR:

• Bulk loads balances via GetBalances() instead of per-validator GetBalance (validators and balances are index-keyed and dense, so balances[i] == validator i).
• Drop the redundant ValidatorIndexByPubkey lookups in processEffectiveBalanceUpdates and processRegistryUpdates, since the loop index already is the validator index.
• Adds temporary per-step timing logs in processEpoch to confirm which step dominates (we can remove this devnet-core testing).

Going to push a tag from this PR and deploy it to devnet-core and monitor

[coderabbitai]

Caution

Review failed

An error occurred during the review process. Please try again later.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch imporove-perf-epoch-deposit

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}