Reward is enough for convex MDPs