Adaptive LoRA Experts Allocation and Selection for Federated Fine-Tuning