Plug-and-Adapt: Multimodal Coreference Resolution at First Sight with a Pretrained Alignment Model