No Preference Left Behind: Group Distributional Preference Optimization