Learning to Ask Informative Questions: Enhancing LLMs with Preference Optimization and Expected Information Gain