Turning Logic Against Itself : Probing Model Defenses Through Contrastive Questions