fix(replication): propagate per-event expiresAt to record writes

[kriszyp]

Problem

Records received via replication were not being evicted by the cleanup scanner even when they carried an expiresAt value. Three issues (same as #639 which backports this to v5.0):

1. expiresAt missing from options passed to _writeUpdate

In a multi-record transaction batch, every event after the first is processed with txnInProgress (the first event) as context. Subsequent events had no expiresAt in their options, so _writeUpdate saw context.expiresAt from the first record applied to all records — or saw nothing if the first record had no expiration.

2. _writeUpdate didn't read expiresAt from options

// before
const expiresAt: number = context?.expiresAt ?? (expirationMs ? expirationMs + Date.now() : -1);

// after
const expiresAt: number = options?.expiresAt ?? context?.expiresAt ?? (expirationMs ? expirationMs + Date.now() : -1);

3. scheduleCleanup() not armed for replicated writes

// before
if (context.expiresAt) scheduleCleanup();

// after
if (expiresAt >= 0) scheduleCleanup();

See also

• Backport to v5.0: fix(replication): propagate per-event expiresAt to record writes #639
• Upstream v4 table copy encoding fix: harperdb/harperdb#3120

---

Signed-off by Claude

[claude]

Missing test for the new options.expiresAt path

What: options?.expiresAt ?? is a new first-priority branch. In the replication loop it's populated from event.expiresAt (line 331); for local writes the third arg to the static put call flows into context.expiresAt via the transactional() wrapper, so every existing test in caching.test.js exercises the context?.expiresAt fallback, not this new leg.

Why it matters: A regression to this path — e.g. silently reverting the resolution order — would not be caught by any existing test. This is specifically the path the PR claims to fix.

Suggested fix: Add a test that calls _writeUpdate with options.expiresAt set but no context.expiresAt, and asserts the stored entry carries the correct expiresAt value (and that the cleanup scanner is armed). The "Handles eviction-only config without expiration" shape in caching.test.js is a reasonable model.

[claude]

1 blocker: the new options?.expiresAt branch in _writeUpdate (line 1705) has no test coverage — existing tests flow expiry through context.expiresAt via the transactional() wrapper, so a regression to the replication path wouldn't be caught. Details in the inline comment.