[fix][test] Fix flaky AuditorBookieTest.testBookieClusterRestart

[lhotari]

Fixes: #21538

Motivation

AuditorBookieTest.testBookieClusterRestart is flaky (issue #21538). Investigating a recent failure showed two independent problems that combine to make the test fail intermittently:

1. Bookie cookie mismatch on restart. The test "restarts" the cluster with stopBKCluster() followed by startBKCluster(). startBKCluster() recreates the bookies with fresh data directories and pulls new ports from PortManager. PortManager can hand back a port a just-stopped bookie used, so a new bookie may come up on a recycled host:port with a different data directory. That fails bookie cookie validation against the cookie still registered in the metadata store, throwing BookieException$InvalidCookieException from startBKCluster():

   org.apache.bookkeeper.bookie.BookieException$InvalidCookieException: Cookie [...
   bookieHost: "127.0.0.1:42989"
   journalDir: "/tmp/bookie14854439573575621802test" ...]
   is not matching with [...
   bookieHost: "127.0.0.1:42989"
   journalDir: "/tmp/bookie10380846176335766468test" ...]
   

2. Stale auditor electors leaking across runs. stopAuditorElectors() shuts the electors down but never clears the auditorElectors map. The same test instance is reused across test methods (and across TestNG retries), so a later run that iterates over auditorElectors observes stale, already-stopped electors whose isRunning() returns false, tripping the assertion:

   java.lang.AssertionError: Auditor elector is not running\!
       at org.apache.bookkeeper.replication.AuditorBookieTest.testBookieClusterRestart(AuditorBookieTest.java:133)
   

   This is the assertion reported in Flaky-test: AuditorBookieTest.testBookieClusterRestart #21538. It surfaced on a retry: the first attempt failed with the cookie mismatch (1), and the retry — running on the same instance with the previous run's shut-down electors still in the map — failed on this assertion in a few milliseconds.

Modifications

• Restart the cluster with the existing restartBookies() helper, which preserves each bookie's identity (host:port and data directories) so cookie validation passes. This also drops the no-longer-needed metadataServiceUri juggling that startBKCluster() required.
• Clear the auditorElectors map in stopAuditorElectors() so stale, shut-down electors are not re-checked by a subsequent test method or retry.

Verifying this change

This change is already covered by the existing AuditorBookieTest.

It was verified by running the class with @Test(invocationCount = 10) and -PtestRetryCount=0, which reproduces the same-instance retry scenario from CI: all 10 invocations of testBookieClusterRestart pass (0 failures, 0 skipped). spotlessCheck, checkstyleMain, and checkstyleTest pass.

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