RangeViT: Towards Vision Transformers for 3D Semantic Segmentation in Autonomous Driving