Iterative Inference in a Chess-Playing Neural Network