The Scaling Law for LoRA Base on Mutual Information Upper Bound