This makes blockstore teardown more resilient to failures, trying harder
to continue and clean up when errors are encountered.  In particular, we
now try multiple times to delete a busy resource, waiting in between
attempts.