A Survey on Model Extraction Attacks and Defenses for Large Language Models