Use PyThreadState_GET() in performance critical code

It seems like _PyThreadState_UncheckedGet() is not inlined as expected, even
when using gcc -O3.

Python/errors.c

@@ -161,7 +161,7 @@ PyErr_SetString(PyObject *exception, const char *string)
 PyObject *
 PyErr_Occurred(void)
 {
-    PyThreadState *tstate = _PyThreadState_UncheckedGet();
+    PyThreadState *tstate = PyThreadState_GET();
     return tstate == NULL ? NULL : tstate->curexc_type;
 }