Ceph

Right now rados bench write leaves a lot of data just lying around the cluster, without a good way to clean itself up.

• by default, clean up after each run
• optionally, do not clean up
• advertise our object prefix loudly on startup
• make the prefix always share some unique-ish prefix (before the unique-to-this-run-goo)
• allow cleanup of earlier runs by specifying the prefix, or all runs by specifying the generic prefix