e2e(kafka): the backup round trip's pre-clean leaves PVCs from an interrupted run, and its cleanup comment describes ZooKeeper #4649

Two leftovers in the kafka backup round trip after the PVC cleanup for #4616.

The pre-clean step of kafka-2-backup-roundtrip (hack/e2e-chainsaw/kafka/chainsaw-test.yaml) runs cleanup.sh but does not delete broker PVCs an interrupted earlier run left behind. The node pools set deleteClaim: false, so a same-named Kafka created next binds the stale volumes, whose on-disk cluster id does not match, and the brokers crash-loop with InconsistentClusterIdException (the failure the kafka README describes). Fix shape: the same label-selected PVC delete as the suite's own cleanup, right after the pre-clean cleanup.sh, waited on.

The comment above the teardown waits in examples/backups/kafka/cleanup.sh says each Kafka app uninstall drains "two StatefulSets, ZooKeeper and their PVCs". The chart runs KRaft node pools, and the PVCs survive the uninstall on purpose, so both halves are false. Fix shape: describe what the waits actually wait for.