A General Framework for Learning-Based Distributionally Robust MPC of Markov Jump Systems