WindowClone: always pass create_child_func to bind_model

[lenemter]

Fixes #2934

I assume due to some race condition mutter tries to call null as a function even after we call bind_model (null, ...). I reproduced the crash and ClutterActor *child = priv->create_child_func (item, priv->create_child_data);
is where the crash occurs and this branch fixed the crash for me.

static void
clutter_actor_child_model__items_changed (GListModel *model,
                                          guint       position,
                                          guint       removed,
                                          guint       added,
                                          gpointer    user_data)
{
  ClutterActor *parent = user_data;
  ClutterActorPrivate *priv = parent->priv;
  guint i;

  while (removed--)
    {
      ClutterActor *child = clutter_actor_get_child_at_index (parent, position);
      clutter_actor_destroy (child);
    }

  for (i = 0; i < added; i++)
    {
      GObject *item = g_list_model_get_item (model, position + i);
      ClutterActor *child = priv->create_child_func (item, priv->create_child_data);

      /* The actor returned by the function can have a floating reference,
       * if the implementation is in pure C, or have a full reference, usually
       * the case for language bindings. To avoid leaking references, we
       * try to assume ownership of the instance, and release the reference
       * at the end unconditionally, leaving the only reference to the actor
       * itself.
       */
      if (g_object_is_floating (child))
        g_object_ref_sink (child);

      clutter_actor_insert_child_at_index (parent, child, position + i);

      g_object_unref (child);
      g_object_unref (item);
    }
}

[utf-4096]

This doesn’t seem to resolve the crash for me, same backtrace. I added some warning ()s around bind_model and I don't see any logs in journalctl, so I’m not sure this code path is being reached.