Detect use of type vectorcall slot and use fallback implementation

Following the merge of python/cpython#123332,
type slot 82 is reserved for type vector calls. This means that we can
safely intercept this value if provided to nanobind even on older
versions of Python where the feature isn't supported yet. In this case,
the metaclass-based implementation is used.

Author: wjakob

src/nb_type.cpp

@@ -13,9 +13,9 @@
 #  pragma warning(disable: 4706) // assignment within conditional expression
 #endif
 
-// #if !defined(Py_tp_vectorcall)
-// #  define Py_tp_vectorcall 82
-// #endif
+#if !defined(Py_tp_vectorcall)
+#  define Py_tp_vectorcall 82
+#endif
 
 NAMESPACE_BEGIN(NB_NAMESPACE)
 NAMESPACE_BEGIN(detail)
@@ -1118,11 +1118,11 @@ PyObject *nb_type_new(const type_init_data *t) noexcept {
             has_custom_init_or_new |=
                 slot_id == Py_tp_init || slot_id == Py_tp_new;
 
-            // if (slot_id == Py_tp_vectorcall) {
-            //     type_vectorcall = (vectorcallfunc) ts.pfunc;
-            //     has_custom_type_vectorcall = true;
-            //     continue;
-            // }
+            if (slot_id == Py_tp_vectorcall) {
+                type_vectorcall = (vectorcallfunc) ts.pfunc;
+                has_custom_type_vectorcall = true;
+                continue;
+            }
 
             *s++ = ts;
         }