[Test][Zeta] Wait for final canceled state in role test

[DanielLeens]

What does this PR do?

This is a follow-up CI baseline fix split out from #11136.

#11136 hit a new unrelated failure in unit-test (8, ubuntu-latest):
SeaTunnelEngineClusterRoleTest.testWorkerIsFirstMemberThenGetJobDetailStatus

The failure was not in the Pulsar code path. The test was asserting jobClient.getJobStatus(jobId) directly after cancelJob(jobId) during master failover, and CI showed the transient state could remain CANCELING before the final terminal result became visible.

This change makes the test wait for clientJobProxy.waitForJobCompleteV2() first, then asserts the final client-visible status is CANCELED.

Why is this needed?

The previous version only increased timeout and polling, but the failing CI still timed out on the transient CANCELING state. Waiting on the final JobResult matches the real completion boundary and avoids sampling the intermediate state as a terminal assertion.

Validation

Ran:

• ./mvnw -nsu -pl seatunnel-engine/seatunnel-engine-client spotless:apply

Attempted but blocked by unrelated local compile issues already present on the clean apache/dev worktree:

• ./mvnw -nsu -pl seatunnel-engine/seatunnel-engine-client -am -DfailIfNoTests=false -Dtest=SeaTunnelEngineClusterRoleTest#testWorkerIsFirstMemberThenGetJobDetailStatus test
• ./mvnw -nsu -pl seatunnel-engine/seatunnel-engine-client -DfailIfNoTests=false -Dtest=SeaTunnelEngineClusterRoleTest#testWorkerIsFirstMemberThenGetJobDetailStatus test

The local blocker was not this test change itself:

• reactor build path hit missing local seatunnel-config-shade sources in this machine's worktree
• module-only path hit existing MultipleTableJobConfigParserTest testCompile errors on the clean apache/dev checkout before this change could be exercised

seatunnel-engine-ui was not changed and was not revalidated in this run.

[nzw921rx]

+1 LGTM