Add metrics for watermarks recorded in recordWatermark() and watermarks evicted/popped in mergeAndValidate() (code).

The motivation here is a healthy system should record and evict approximately the same number of watermarks. If not, then we aren't seeing successful checkpoints or otherwise too far behind, and we are maintaining a lot of deltas in-memory indefinitely.

See also comment.

 Also add a counter for checkpoints the standby skips because it never recorded their watermark (the continue in mergeAndValidate()), so tests can tell whether a checkpoint was validated or skipped without parsing log lines. A count that keeps growing on a healthy standby also means its watermarks are being lost.