FLoRA: Enhancing Vision-Language Models with Parameter-Efficient Federated Learning