Generative Data Augmentation using LLMs improves Distributional Robustness in Question Answering