Deflake TestActivityLog_MultipleFragmentsAndSegments

[ncabatoff]

Example failure: https://github.com/hashicorp/vault-enterprise/actions/runs/5144577938/jobs/9261153205

Fix is never starting the fragment worker instead of racing on stopping it. I added a log (since removed) so we can see more details about failures:

2023-06-01T09:41:46.769-0400 [TRACE] TestActivityLog_MultipleFragmentsAndSegments.activity: new local fragment created
2023-06-01T09:41:46.769-0400 [TRACE] TestActivityLog_MultipleFragmentsAndSegments.activity: shutting down activeFragmentWorkder

The issue was that by closing doneCh, we were hoping that our read of newFragmentCh wouldn't race with activeFragmentWorker's read. However, we don't control when activeFragmentWorker gets scheduled by the runtime, and sometimes it doesn't get scheduled until the new fragment has been created and newFragmentCh is readable. In that case, there's no guarantee that the doneCh read will happen instead of the newFragmentCh read, and when that happens we get

    activity_log_test.go:762: timed out waiting for new fragment