Multimodal Contrastive Learning with LIMoE: the Language-Image Mixture of Experts