TowardtheFundamentalLimitsofImitation Learning