Know Your Limits: A Survey of Abstention in Large Language Models