A FLEXIBLE GENERAL CLASS OF MARGINAL AND CONDITIONAL RANDOM INTERCEPT MODELS FOR BINARY OUTCOMES USING MIXTURES OF NORMALS