RoCoIns: Enhancing Robustness of Large Language Models through Code-Style Instructions