Flaky TestRegisterAndUnregister due to asynchronous ATS goroutine shutdown

Observed behavior

Running go test -run TestRegisterAndUnregister -count=5 in server/ats/ats_test.go can fail:

--- FAIL: TestRegisterAndUnregister (0.50s)
    ats_test.go:72: Expected same number of go routines after removing all registered: 3 vs 2
--- FAIL: TestRegisterAndUnregister (0.50s)
    ats_test.go:72: Expected same number of go routines after removing all registered: 3 vs 2
FAIL
exit status 1
FAIL    github.com/nats-io/nats-server/v2/server/ats    3.112s

Expected behavior

This test should pass without asssertion failure

Server and client version

v2.14.3

Host environment

Not relevant to the bug.

Steps to reproduce

Running go test -run TestRegisterAndUnregister -count=100 in folder server/ats can reproduce the failure.

The bug is caused by the ATS goroutine shutdown being asynchronous. Unregister() only sends a stop signal to the global done channel, but it does not wait for the ticker goroutine to actually receive the signal and exit. As a result, when Register() and Unregister() are called repeatedly or concurrently, an old ticker goroutine may still be alive while a new one is started. This can temporarily leave multiple ATS ticker goroutines running, and in tests such as TestRegisterAndUnregister -count=100, a leftover goroutine from the previous iteration can be counted by runtime.NumGoroutine(), causing flaky failures.