Summary

The producer's rolling previous_journal_hash was an in-memory Mutex<B256> placeholder. This PR persists the keccak of each block's wire-encoded journal alongside the block in hot storage and seeds the next emit from there, making the rolling hash crash- and revert-safe.

Changes

• encode_journal now stamps ExecutedBlock::journal_hash, which UnifiedStorage::append_blocks persists into the new JournalHashes hot table; previous_journal_hash reads it back via HotDbRead::get_journal_hash (falling back to GENESIS_JOURNAL_HASH only for empty storage or pre-JournalHashes upgrades, and only while the in-process chain is itself fresh).
• Per-block ordering changed to emit -> append -> notify, so a post-emit failure leaves storage at N-1 and a fresh-chain restart re-emits N deterministically against tip = None; notify last so eth_subscribe("newHeads") clients querying storage immediately after the broadcast see block N already indexed.
• on_host_revert now returns a RevertOutcome { changed, shutdown }: when the post-revert target is genesis or has been evicted from the chain's ring buffer, storage and tags are drained, the post-notification tag refresh still runs (so reth gets a FinishedHeight matching post-drain state), then the node bails with an explanatory error. A tag-refresh failure on the shutdown path is logged via error! so it can't mask the shutdown error.
• Removed the #[ignore = "ENG-2017: ..."] markers from the reorg tests in multiple-blocks.rs and reorg.rs; reorg tests that previously reverted to height 0 now reorganize down to height 1 instead, since the new code bails on a true revert-to-genesis.
• Added test_journal_hash_persisted_after_process_block (db.rs) covering the persistence write, and test_revert_to_genesis_bails (basic.rs) covering the new bail path.

Notes

• Depends on feat(hot)!: add table for journal hashes storage#70; patched in [patch.crates-io] until the storage repo is published.
• Linear: ENG-2017.

🤖 Generated with Claude Code