Javadoc
Find the earliest commit log position that is not covered by the known flushed ranges for some table.
For efficiency this assumes that the first contiguously flushed interval we know of contains the moment that the
given table was constructed* and hence we can start replay from the end of that interval.
If such an interval is not known, we must replay from the beginning.
* This is not true only until if the very first flush of a table stalled or failed, while the second or latter
succeeded. The chances of this happening are at most very low, and if the assumption does prove to be
incorrect during replay there is little chance that the affected deployment is in production.