Improving precision by adjusting for baseline variables in randomized trials with binary outcomes, without regression model assumptions