Widen flaky poll timeouts in maintenance, STT, and instance-op integration tests

[sjainit]

Issues

• My PR addresses the following Helix issues and references them in the PR description:

Fixes several pre-existing flaky integration tests that intermittently fail in CI with expected:<true> but was:<false>, unrelated to the change under test. Most recently testEvacuateWithDisabledPartition failed this way on a PR CI run (Tests run: 1846, Failures: 1). No tracking issue is currently filed — happy to open one if preferred.

Description

• Here are some details about my PR:

What. Test-only change that stabilizes three flaky integration-test classes by making their convergence polls wait long enough and assert the right condition.

Why. Each test polls for cluster convergence and then asserts on the outcome, but the polls use tight/default timeouts (30s) or — in the maintenance test — wait on a different condition than the one asserted. Under CI scheduling contention a poll can return before the cluster converges, yielding an intermittent AssertionError: expected:<true> but was:<false> that has nothing to do with the change under test. For example testEvacuateWithDisabledPartition polled _admin.isEvacuateFinished(...) for only 30s; under load evacuation took longer (elapsed 36.8s) and the assert flaked.

How.

• TestClusterMaintenanceMode — re-point 13 polls at the exact asserted condition (maintenance-signal presence/absence, specific live-instance names, or the history entry's OPERATION_TYPE), assert the poll result instead of ignoring it, and wait TestHelper.WAIT_DURATION. Some polls previously waited on stale/wrong live-instance counts; these are corrected to count-independent or name-based predicates.
• TestStateTransitionTimeout — assert the verifyByPolling result and use the 60s overload instead of the default 30s.
• TestInstanceOperation — widen five 30s verifier(...) polls to TestHelper.WAIT_DURATION.

Widening these budgets is safe: each poll returns as soon as its predicate becomes true, so a passing run is unaffected; only slow-but-correct runs get the extra time they need.

Tests

• Test-only PR (no product code touched). Changed classes:

TestClusterMaintenanceMode, TestStateTransitionTimeout, TestInstanceOperation.

• mvn test result: TestClusterMaintenanceMode and TestStateTransitionTimeout were validated locally under a CPU-throttle reproduction harness — both pass 3/3 iterations at the load that previously reproduced the flake, and pass at normal speed with no regression. TestInstanceOperation compiles cleanly but is not locally reproducible in isolation (deep dependsOnMethods chain); relying on CI for that class.

Changes that Break Backward Compatibility (Optional)

None. Test-only change; no production code, public API, config, or serialization-format changes.

Documentation (Optional)

None.

Commits

• Commit subject uses the imperative mood; the body explains what and why (not how) and is wrapped.

Code Quality

• Diff follows the helix-style formatting conventions.