fix(stdlib): prevent block hash deletion for blocks still referenced by tips

[AztecBot]

Summary

Fixes a crash loop in the p2p block stream caused by handleChainFinalized in L2TipsStoreBase deleting block hashes that are still referenced by other tips.

Root Cause

handleChainFinalized called deleteBlockHashesBefore(event.block.number) which could delete hashes for blocks still referenced by other tips (e.g. proposedCheckpoint, proven). When getL2Tips() subsequently tried to look up the hash for those tips via getBlockId(), it threw Block hash not found for block number N, putting the p2p block stream into an unrecoverable error loop that prevented block sync and caused the e2e_bot test to timeout.

Fix

Compute a safe deletion bound as the minimum of all active tip block numbers before deleting. This ensures no block hash is deleted while it's still referenced by any tip.

Test plan

• All 45 l2_block_stream tests pass
• All 207 archiver KV store tests pass
• Full yarn-project build succeeds

ClaudeBox log: https://claudebox.work/s/4aa87310c1ecd68c?run=1

[AztecBot]

Automatically closing this stale claudebox draft PR (no updates for 5+ days). Re-open if still needed.

[guglez]

We hit this exact bug on our mainnet validator running v4.2.1. The node stopped attesting silently - the container kept running, archiver and HTTP API were healthy, but the sequencer entered an internal error loop at ~2 errors/second (Block hash not found for block number 46069). No crash, no obvious signal. Recovery: stop container, delete the p2p LMDB stores (p2p, p2p-archive, p2p-peers, p2p-attestation), restart. Node came back immediately. Any ETA on a release with this fix included?

[aminsammara]

We hit this exact bug on our mainnet validator running v4.2.1. The node stopped attesting silently - the container kept running, archiver and HTTP API were healthy, but the sequencer entered an internal error loop at ~2 errors/second (Block hash not found for block number 46069). No crash, no obvious signal. Recovery: stop container, delete the p2p LMDB stores (p2p, p2p-archive, p2p-peers, p2p-attestation), restart. Node came back immediately. Any ETA on a release with this fix included?

Hi @guglez looking into this

[spalladino]

Hey @guglez, we fixed an instance of this issue on the upcoming V5 release in #23295. Can you share some logs, to confirm your error is indeed the same and would be fixed by that?

[spypsy]

relevant fix now in #23505

[guglez]

Hey @spalladino

Is this log line enough?

  ERROR: kv-store:lmdb-v2 Failed to commit transaction: Error: Block hash not found for block number 46069
      at L2TipsKVStore.getBlockId (file:///usr/src/yarn-project/stdlib/dest/block/l2_block_stream/l2_tips_store_base.js:76:19)
      at async Promise.all (index 2)
      at async file:///usr/src/yarn-project/stdlib/dest/block/l2_block_stream/l2_tips_store_base.js:14:93