[fix][test] Fix flaky BacklogQuotaManagerTest.createNamespaces (HTTP 409 cascade)

[lhotari]

Motivation

When CI runs BacklogQuotaManagerTest, two cases fail together as a cascade:

1. BacklogQuotaManagerTest.clearNamespaces (the @AfterMethod) fails with org.awaitility.core.ConditionTimeoutException: Condition ... was not fulfilled within 10 seconds. while force-deleting prop/ns-quota.
2. The next test method's @BeforeMethod createNamespaces then fails with PulsarAdminException$ConflictException: Namespace already exists (HTTP 409), because the prior cleanup never finished — and the same 409 keeps cascading for every subsequent test.

Example: https://github.com/apache/pulsar/actions/runs/25387367872/job/74455105884 (PR #25676 attempt 1; the failure is unrelated to that PR's change).

Root cause: PR #25624 added testConsumerBacklogEvictionSizeQuotaCleansPendingAcks and testConsumerBacklogEvictionTimeQuotaNotPreciseCleansPendingAcks, which leave Key_Shared consumers with unacked messages on prop/ns-quota and an active backlog-quota check loop running every 2s. Force-deleting that state can exceed the 10s default Awaitility budget used by MockedPulsarServiceBaseTest.deleteNamespaceWithRetry. The afterMethod times out, leftover metadata remains, and the next beforeMethod hits 409.

Modifications

1. MockedPulsarServiceBaseTest.deleteNamespaceWithRetry now sets atMost(20, TimeUnit.SECONDS) instead of relying on Awaitility's 10s default. 2× headroom is enough for the heaviest cleanups while staying conservative for the many tests that share this helper.
2. BacklogQuotaManagerTest.createNamespaces is split: a new createNamespaceForTest(String) helper catches PulsarAdminException.ConflictException and recovers by force-deleting the leftover and recreating. Defense in depth — even if cleanup ever exceeds the 20s budget, the next test starts with clean state instead of cascading 409s.

Verifying this change

• Make sure that the change passes the CI checks.

This change is already covered by existing tests, such as BacklogQuotaManagerTest.testConsumerBacklogEvictionSizeQuotaCleansPendingAcks and BacklogQuotaManagerTest.testConsumerBacklogEvictionTimeQuotaNotPreciseCleansPendingAcks (the two heavy tests that exposed the cascade). Verified locally with @Test(invocationCount = 10) on both: 20/20 passes, all clearNamespaces cycles within the new 20s budget.

Does this pull request potentially affect one of the following parts:

• Dependencies (add or upgrade a dependency)
• The public API
• The schema
• The default values of configurations
• The threading model
• The binary protocol
• The REST endpoints
• The admin CLI options
• The metrics
• Anything that affects deployment