From Understanding to Utilization: A Survey on Explainability for Large Language Models