add a check that ensures we see the number of shards that we would expect in the cluster state before the test starts and retry after a pause if we do not