Readless increments can drop cells on flush and compaction

The readless increment implementation can drop increment cells when a region flush or compaction occurs. This is due to 2 separate problems:

• when a limit is specified to the scanner's next() call, we fill a list with results from the underlying scanner, but we may not process all the results in the list if we hit the limit prior to handling all entries in the current batch; and

• the Tephra TransactionProcessor coprocessor sets a version limit on the store scanner that it uses, which can prematurely cut off the results before all increment cells are seen (see TEPHRA-49)