work out the data.frame column types during parsing

rather than after-the-fact inside simplify_dataframe(). If we can do it during the main json_to_sexp() loop we should save some time.