Mediated Uncoupled Learning and Validation with Bregman Divergences: Loss Family with Maximal Generality