Fed-piLot: Optimizing LoRA Assignment for Efficient Federated Foundation Model Fine-Tuning