Privately Learning High-Dimensional Distributions