Effects of Distributional Biases on Gradient-Based Causal Discovery in the Bivariate Categorical Case