Rethinking Prototypical Contrastive Learning through Alignment, Uniformity and Correlation