LIV: Language-Image Representations and Rewards for Robotic Control