CLAMP-ViT: Contrastive Data-Free Learning for Adaptive Post-Training Quantization of ViTs