Stop RecordingApmServer message processing before returning from tests

[ldematte]

The RecordingApmServer needs to be started before the ElasticsearchCluster; however, it needs to be stopped (or, at least, it needs stopping processing messages) before the ElasticsearchCluster.
JUnit rules do process before/after in reverse order at startup/shutdown; using a RuleChain slightly improves the situation (as opposed to today's ClassRule + Rule), but does not solve the issue.
Here I am introducing an additional "stop" method; the method needs to be called explicitly at the end of the test, but this way we can make RecordingApmServer stop processing requests before the ElasticsearchCluster is gone and communication is interrupted exceptionally.

Fixes #129651

[elasticsearchmachine]

Pinging @elastic/es-core-infra (Team:Core/Infra)

[mosche]

lgtm, though wondering if it wouldn't be easier to reason if instead handling IOExceptions

[rjernst]

I'm confused how this addresses the linked failure. I thought the problem was the apm agent continues to run in the background, so requests to the recording server begin failing. Can you explain how this change fixes the issue?

[ldematte]

Turned out that the issue is that the APM agent is in the middle of writing, ES gets shut down, the connection breaks, the mock APM server throws.
Shutting down the APM server before ES brings things down cleanly.
But thinking again, this is still brittle and an implementation detail; probably catching all exceptions like we were discussing is still better (more robust and simpler)

[ldematte]

@mosche I've decided to go with your suggestion; I tried to refactor the RecordingApmServer to better coordinate the APM mock server and the ES cluster, but it's error prone and I was not satisfied. So better to go with the minimalist and more robust solution, at least until we stay with the APM agent.

[mosche]

lgtm 👍