Exploring Robustness of Multilingual LLMs on Real-World Noisy Data