ProvableRepresentationLearningforImitation withContrastiveFourierFeatures