Privately Learning Decision Lists and a Differentially Private Winnow