Using Natural Language Explanations to Improve Robustness of In-context Learning for Natural Language Inference