Synthesis of separation processes with reinforcement learning