One-shot Imitation Learning via Interaction Warping