Overcoming Hallucinations in AI: How Factually Augmented RLHF Optimizes Vision-Language Alignment in Large Multimodal Models