fix: make gil_safe_call_once thread-safe in free-threaded CPython (#5246)

* fix: Make gil_safe_call_once thread-safe in free-threaded CPython

The "is_initialized_" flags is not protected by the GIL in free-threaded
Python, so it needs to be an atomic field.

Fixes #5245

* style: pre-commit fixes

* Apply changes from code review

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

include/pybind11/gil_safe_call_once.h

@@ -8,6 +8,10 @@
 #include <cassert>
 #include <cassert>
 #include <mutex>
 #include <mutex>
 
 
+#ifdef Py_GIL_DISABLED
+#    include <atomic>
+#endif
+
 PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 
 
 // Use the `gil_safe_call_once_and_store` class below instead of the naive
 // Use the `gil_safe_call_once_and_store` class below instead of the naive
@@ -82,7 +86,12 @@ public:
 private:
 private:
     alignas(T) char storage_[sizeof(T)] = {};
     alignas(T) char storage_[sizeof(T)] = {};
     std::once_flag once_flag_ = {};
     std::once_flag once_flag_ = {};
-    bool is_initialized_ = false;
+#ifdef Py_GIL_DISABLED
+    std::atomic_bool
+#else
+    bool
+#endif
+        is_initialized_{false};
     // The `is_initialized_`-`storage_` pair is very similar to `std::optional`,
     // The `is_initialized_`-`storage_` pair is very similar to `std::optional`,
     // but the latter does not have the triviality properties of former,
     // but the latter does not have the triviality properties of former,
     // therefore `std::optional` is not a viable alternative here.
     // therefore `std::optional` is not a viable alternative here.