Optimal Transport for Offline Imitation Learning