The test had two flaky assertions:
1. Strict equality check that followers are exactly 1 commit offset
   behind the leader - replaced with assert.Eventually that waits for
   followers to catch up
2. Point-in-time checksum comparison across replicas without waiting
   for follower replication to complete - replaced with assert.Eventually
   that waits for follower checksums to match the leader

Also increased assert.Eventually timeouts from 10s to 30s for
robustness under CI load.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The test was flaky because commit notifications are piggybacked on
replication messages, so followers never learn the last committed
entry's status without a subsequent write.

- Write all keys before checking feature enabled (flushes the
  feature-enable commit notification to followers)
- Record leader checksum, then write a flush entry to propagate
  the commit notification for the last key
- Wait for followers to reach the recorded commit offset
- Compare checksums with assert.Equal for clear error messages

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>