src: fix perfetto session reader teardown race

[MarshallOfSound]

PerfettoSessionReader::Deleter does a final ReadTrace() then Stop(). Perfetto runs the read callback and the stop callback as separate tasks on its own thread, so the stop can win: the loop thread closes the handles and deletes the reader while a ReadTraceCallback bound to the raw this is still queued, which then locks a destroyed mutex and signals a closed uv_async_t.

This defers teardown until the owner has released the reader, the session has stopped, and no read is in flight, and makes both Perfetto-thread callbacks flip their flag and signal under chunks_mutex_ so the loop thread can't free the reader in between.

Found with ASan (trace_events.createTracing().enable() + process exit, intermittently abort() in uv_mutex_lock from ReadTraceCallback).

Refs: #64565

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.06%. Comparing base (9f04fcd) to head (aeccc1d).
⚠️ Report is 5 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #65611      +/-   ##
==========================================
- Coverage   90.07%   90.06%   -0.01%     
==========================================
  Files         751      751              
  Lines      254916   254916              
  Branches    48133    48121      -12     
==========================================
- Hits       229605   229586      -19     
- Misses      16496    16497       +1     
- Partials     8815     8833      +18     

see 36 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[legendecas]

I think a simple fix could be:

  // Shutdown requested and no read outstanding.
  bool should_tear_down = false;
  {
    Mutex::ScopedLock lock(reader->chunks_mutex_);
    std::swap(chunks_to_write, reader->pending_chunks_);
    should_tear_down =
        reader->stop_requested_ && !reader->read_in_progress_;
  }

and later check it

  if (!should_tear_down || reader->handles_pending_close_ != 0) return;

  ... close handles.

With the Mutex::ScopedLock changes in both ReadTraceCallback and SessionStopCallback.

This does not need any additional mutexes or flags.