Operations can block on cache eviction at any time, and we should ensure that after allocating an oplog slot, we cannot block on cache eviction for longer than some maximum threshold. We otherwise risk delaying replication for extended periods of time.

We can use the setCacheMaxWaitTimeout for this purpose. I suspect this API has a bit of overhead due to the WT_SESSION::reconfigure() call, so we'll need to investigate that tradeoff.