A Practical, Progressively-Expressive GNN