Adapting Vision-Language Models Without Labels: A Comprehensive Survey