Maximum entropy RL (provably) solves some robust RL problems