It is a wasted effort to run eviction on an ingest table if we haven't picked up a new checkpoint since the last eviction attempt.

We need to tune our eviction algorithm to avoid doing eviction and wasting CPU in this case.