Quiet TSAN warnings about remaining non-atomic accesses of `tstate->s…

…tate`

These should be the last instances of TSAN warning of data races when accessing
`tstate->state` non-atomically.

TSAN reports a race between accessing `tstate->state` in `_PyThreadState_Suspend()`
and the compare/exhange in `park_detached_threads`. This is a false positive due
to TSAN modeling failed compare/exchange operations as writes.

TSAN reports a race between accessing `tstate->state` in `_PySemaphore_Wait()`
and the compare/exchange in `park_detached_threads`. This is the same issue that
we saw in gh-117830.

Python/parking_lot.c

@@ -194,7 +194,8 @@ _PySemaphore_Wait(_PySemaphore *sema, PyTime_t timeout, int detach)
     PyThreadState *tstate = NULL;
     if (detach) {
         tstate = _PyThreadState_GET();
-        if (tstate && tstate->state == _Py_THREAD_ATTACHED) {
+        if (tstate && _Py_atomic_load_int_relaxed(&tstate->state) ==
+                          _Py_THREAD_ATTACHED) {
             // Only detach if we are attached
             PyEval_ReleaseThread(tstate);
         }

Python/pystate.c

@@ -2096,7 +2096,7 @@ _PyThreadState_Suspend(PyThreadState *tstate)
 {
     _PyRuntimeState *runtime = &_PyRuntime;
 
-    assert(tstate->state == _Py_THREAD_ATTACHED);
+    assert(_Py_atomic_load_int_relaxed(&tstate->state) == _Py_THREAD_ATTACHED);
 
     struct _stoptheworld_state *stw = NULL;
     HEAD_LOCK(runtime);