Build a NumPy ResNet block with padded convolutions, a skip connection, and ReLU after the residual addition.
