SStablesplit behavior changed

The dtest sstablesplit_test.py has begun failing due to an incorrect number of sstables being created after running sstablesplit.

http://cassci.datastax.com/job/cassandra-2.1_dtest/559/changes#detail1
is the run where the failure began.

In 2.1.x, the test expects 7 sstables to be created after split, but instead 12 are being created. All of the data is there, and the sstables add up to the expected size, so this simply may be a change in default behavior. The test runs sstablesplit without the --size argument, and the default has not changed, so it is unexpected that the behavior would change in a minor point release.