Bending Reality: Distortion-aware Transformers for Adapting to Panoramic Semantic Segmentation