Enhancing Hindi–English Direct Speech-to-Speech Translation with Clustering-Aided Cross-Contrastive Self-Supervised Speech Representation Learning