LocalViT: Analyzing Locality in Vision Transformers