Clustering predicted structures at the scale of the known protein universe