Learning Utilities and Equilibria in Non-Truthful Auctions