MultiWay-Adapater: Adapting large-scale multi-modal models for scalable image-text retrieval