Location-Aware Self-Supervised Transformers for Semantic Segmentation