ULIP: Learning a Unified Representation of Language, Images, and Point Clouds for 3D Understanding