[Tests] Add TestNG listeners for resource cleanup, thread leak detection and "fail fast"

[lhotari]

Motivation

There are a lot of resource cleanup issues in Pulsar tests. This PR adds some TestNG listeners to mitigate the problem.

Modifications

• add TestNG listeners for cleaning up

  • Mockito mocking state in all threads
  • FastThreadLocal state holding org.apache.pulsar.* class instances in all threads

• add TestNG listener for detecting leaked threads

  • This helps noticing and fixing issues caused by threads that are kept running after the test completes

• rewrite RetryAnalyzer based on TestNG's RetryAnalyzerCount class

  • add tests for RetryAnalyzer

• Fix test dependencies in pulsar-io/flume and pulsar-io/netty

• add custom fail fast solution

  • Maven Surefire built-in solution is broken with TestNG 7.3.0
  • This makes the test retries in CI faster since most test runs are retries 3 times. It's a waste of resources to keep executing after the first test failure. (This applies at least for the 2 first runs out of the maximum of 3 runs. It could be useful to disable "fail fast" for the last attempt, however that is out-of-scope of this PR.)

[lhotari]

/pulsarbot run-failure-checks

[eolivelli]

LGTM

@codelipenghui @congbobo184 PTAL
CI passed

[codelipenghui]

👍