Generalizing vision-language models to novel domains: A comprehensive survey