Rate-Distortion Theoretic Bounds on Generalization Error for Distributed Learning