RLIF: Interactive Imitation Learning as Reinforcement Learning