TriCoLo: Trimodal Contrastive Loss for Text to Shape Retrieval