Using Whole Slide Image Representations from Self-Supervised Contrastive Learning for Melanoma Concordance Regression