We have internalSamplingSizeOverride as well as internalJoinPlanSamplingSize. On the analyze path we only consult internalSamplingSizeOverride, but on the read path we may consult either depending on the query. This is confusing so we should write a test that documents this behavior. The test should ensure that both queries destined for CBR and queries destined for join optimization can use a persisted sample if one exists.

This is in anticipation of removing the ability to specify sample size from the analyze command, as well as the anticipation of automated samples.