AMQ-XX: Fix long running KahaDBOffsetRecoveryListenerTest on CI

[jeanouii]

This is taking about an hour to run.
The @before code is running every single time for all the test scenario of the class.

I'm questioning what the benefit of the @before peace of code. If it's an actual test, I'd pull it into its own @test method.
But for now, it's burning a lot of time because it does a lot of IO and the CI is slow for IO. Moreover, the directory used in the sendMessage call of the @before portion is the default one, because the broker is lazily created after using the VM Transport, whereas all the other tests are actually using a different directory.

Locally on my Mac M1 it runs on 16 min and after the removal, it runs in 2'30.

If everyone is happy with it, I can create a JIRA and rename this PR and update the commit message

[jeanouii]

[INFO] Tests run: 8, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 522.1 s -- in org.apache.activemq.store.kahadb.KahaDBOffsetRecoveryListenerTest

Build is green and the test went from an hour, to less than 10 minutes.
@jbonofre Should I create a JIRA for it and update the PR + commit message?