Self-Supervised Contrastive Learning for Unsupervised Phoneme Segmentation