We should figure out whether it's reasonable to enforce timestamp ordering of updates (i.e. that updates with a lower timestamp than a subsequent one on the same key can't be committed). One example where we don't enforce this is test_timestamp14.

This causes issues for durable history especially in the case of timestamped deletes because the stop time pair of a cell may be earlier than its start time pair. The order of these updates also won't be preserved in eviction when we write to the history store so these updates that were once "hidden" will suddenly become visible as a result of eviction.