Skip-Attention: Improving Vision Transformers by Paying Less Attention