Efficient Statistics, in High Dimensions, from Truncated Samples