ViNG: Learning Open-World Navigation with Visual Goals