Text this: Upsampling DINOv2 Features for Unsupervised Vision Tasks and Weakly Supervised Materials Segmentation