Modelling the loss given default distribution via a family of zero-and-one inflated mixture models