Contrastive learning, multi-view redundancy, and linear models