Optimal Regularization Under Uncertainty: Distributional Robustness and Convexity Constraints