Adaptive Human-LLMs Interaction Collaboration: Reinforcement Learning driven Vision-Language Models for Medical Report Generation