[test] Tolerate removed tablet servers in FlussClusterExtension wait helpers

### Search before asking

- [x] I searched in the [issues](https://github.com/apache/fluss/issues) and found nothing similar.

### Description

Follow-up to #3667: the same NPE race exists at three more getTabletServerById call sites in FlussClusterExtension. During a failover the lookup can return null (leader temporarily NO_LEADER, or the server already removed), and since retry(...) only retries AssertionError, the NPE fails the test instead of letting the wait continue.

### Willingness to contribute

- [x] I'm willing to submit a PR!

Start in FlussClusterExtension and locate the three getTabletServerById call sites described in the issue; compare their retry(...) usage with the fix discussed in #3667. Verify that a null lookup during failover does not cause an NPE and that the wait continues until the expected condition is met.

Written by the indexing model from the issue text.