GreedyViG: Dynamic Axial Graph Construction for Efficient Vision GNNs