Towards Concise and Adaptive Thinking in Large Reasoning Models: A Survey