[DavidCockerill]

📝 Summary

Adds a two-node test for HarperFast/harper#4391. That fix stops a replica from using an earlier row's expiry in the same transaction (HarperFast/harper#4390). The wrong expiry made the replica hide the row while the origin kept it. This PR also bumps core to that fix, so the test runs against it.

Closes #1399

🔧 Changes

• core moves to harper#4391's head 57b90174d. It contains main's pin 4f20ed583, so nothing rolls back. Once harper#4391 merges, core moves to its merge commit.
• The bump changes which expiry a replicated write takes, not how the receiver groups writes into transactions. harper#4391 states its per-write cost.

✅ Verification

• replicatedWriteOwnExpiry.test.mjs, with its fixture (schema, two one-transaction resources, config): one request writes Exp (expiration: 5) and then Plain in one transaction, another the reverse. Once Exp expires on B, B still has both Plain rows. With core at main's pin, B loses the row written after Exp; with this core it passes in every run.
• Gates: typecheck, lint:required and format pass. test:unit has 3 failures, all identical with core at main's pin (/proc and the /var symlink on macOS). The new test, addNodeTableExpiration and cacheReplicationSource pass together locally; CI runs the full cluster and non-cluster tiers.
• Performance: the bumped core change runs on every replicated write. A write with no expiry in its options or context reads what it read before; only one that would take the context's expiry reads one more property. No allocation, lock or await, so no benchmark (harper#4391).
• After review: the expiry wait now also requires the Exp rows on B first, and a read error counts as neither 0 nor 1 (the final round's finding), not re-reviewed.
• PR QA: skipped @ b87c394 — not substantive: a test and a core bump whose behavior change is harper#4391's

_{🤖 Generated by Claude Code (Anthropic Claude Opus); posted via @DavidCockerill.}

Release target: v5.4 — ships with harper#4391's fix, milestoned v5.4

Depends-on: HarperFast/harper#4391

Related PRs: #1395 independent (the core sync this branch builds on)
Complexity: easy

Deferred findings

• 20189ba273efa47b minor performance — core — No Performance statement for per-write expiry selection on replication apply path brought in by core bump — NOT FILED
• 7d37570d95d95fb8 nit data-integrity — core — Fix might split an origin transaction into per-event replica transactions, breaking crash atomicity — NOT FILED

_{Review-Coverage: authored=claude; ran=gemini,codex,gemini-carbon; late=gemini-carbon:adjudicated; adjudicated=domain; declined=cursor-grok,cursor-composer,muse; rounds=2; full=2 @ b87c394}

_{Review-Attention: skim ~2m (decisions: cluster-test-vs-unit-only) @ b87c394}