Use TaskTracket in streamer for graceful exit and avoiding flaky tests

[KirillLykov]

Problem

When streamer service is launched, it spawns task and returns handle for this task.
In tests we typically do the following:

exit.store(true, Relaxed);
task_handle.await;

//check some stats

Yet inside of this root task of streamer, it creates many other tasks to handle connections and streams, which this root task has no idea about.
Which means that when the root task has stopped we have no idea if it's child tasks have completed or not.

This leads to flaky tests. For example, in nonblocking::quic::test::test_quic_server_multiple_streams we stop the streamer and after that check stats.total_connections.load(Ordering::Relaxed). This counter counts how many task are handling at giving moment connections. It might happen that these tasks haven't finished before we do this check, so all the checks below task_handle.await are potentially flaky.

Summary of Changes

Use TaskTracker as tokio documentation suggests

[KirillLykov]

localnet failure is due to exit timeout. I suspect that this is due to we rely on runtime to shutdown instead of proper cancelling all the tasks because now we wait for all the tasks to finish first. This is fixed in #8025
Alternatively, I could have ignored these tasks in the spawn_server but instead use it only in tests.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 96.77419% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 83.0%. Comparing base (b6c922c) to head (ee7d17a).
⚠️ Report is 2082 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff            @@
##           master    #8066     +/-   ##
=========================================
- Coverage    83.0%    83.0%   -0.1%     
=========================================
  Files         826      826             
  Lines      362561   362581     +20     
=========================================
+ Hits       301086   301100     +14     
- Misses      61475    61481      +6     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[KirillLykov]

@lijunwangs @alexpyattaev this PR is ready for review.

[lijunwangs]

The changes itself look good. Do we have any performance bench with/without this and see if any difference we should worry about as this related to the core of tracking tasks?

[alexpyattaev]

LGTM. TaskTracker is an AtomicUsize, it should not impact perf.

[KirillLykov]

The changes itself look good. Do we have any performance bench with/without this and see if any difference we should worry about as this related to the core of tracking tasks?

It is

struct TaskTrackerInner {
    state: AtomicUsize,
    on_last_exit: Notify,
}

So it is really lightweight and hardly this will lead to any observable performance effect.

[lijunwangs]

LGTM