Avoid thread termination in scoped_released

Do not call `PyEval_RestoreThread()` from `~gil_scoped_release()` if python runtime is finalizing, as it will result in thread termination in Python runtime newer than 3.6, as documented in https://docs.python.org/3/c-api/init.html#c.PyEval_RestoreThread
Similarly do not call `PyThreadState_DeleteCurrent` from `~gil_scoped_acquire()` if runtime is finalizing.

Discovered while debugging PyTorch crash using Python-3.9 described in  pytorch/pytorch#47776

Author: malfet

include/pybind11/pybind11.h

@@ -2121,7 +2121,12 @@ class gil_scoped_acquire {
                     pybind11_fail("scoped_acquire::dec_ref(): internal error!");
             #endif
             PyThreadState_Clear(tstate);
+#if (PY_MAJOR_VERSION == 3 && PY_MINOR_VERSION > 6) && !defined(Py_LIMITED_API)
+            if (!_Py_IsFinalizing())
+                PyThreadState_DeleteCurrent();
+#else
             PyThreadState_DeleteCurrent();
+#endif
             PYBIND11_TLS_DELETE_VALUE(detail::get_internals().tstate);
             release = false;
         }
@@ -2153,7 +2158,14 @@ class gil_scoped_release {
     ~gil_scoped_release() {
         if (!tstate)
             return;
+#if (PY_MAJOR_VERSION == 3 && PY_MINOR_VERSION > 6) && !defined(Py_LIMITED_API)
+        // PyEval_RestoreThread() should not be called if runtime is finilizing
+        // See https://docs.python.org/3/c-api/init.html#c.PyEval_RestoreThread
+        if (!_Py_IsFinalizing())
+            PyEval_RestoreThread(tstate);
+#else
         PyEval_RestoreThread(tstate);
+#endif
         if (disassoc) {
             auto key = detail::get_internals().tstate;
             PYBIND11_TLS_REPLACE_VALUE(key, tstate);