Bivariate Causal Discovery Using Rate-Distortion MDL: An Information Dimension Approach