LViT: Language Meets Vision Transformer in Medical Image Segmentation