SA-FedLora: Adaptive Parameter Allocation for Efficient Federated Learning with LoRA Tuning