Robust Reinforcement Learning under Diffusion Models for Data with Jumps