Model produced by SdcaBinaryTrainer is not and can not be strongly-typed

[wschin]

The CreatePredictor function,

        using TScalarPredictor = IPredictorWithFeatureWeights<float>;
        ...
        protected override TScalarPredictor CreatePredictor(VBuffer<float>[] weights, float[] bias)
        {
            ...
            var predictor = new LinearBinaryModelParameters(Host, in maybeSparseWeights, bias[0]);
            if (Info.NeedCalibration)
                return predictor;
            return new ParameterMixingCalibratedPredictor(Host, predictor, new PlattCalibrator(Host, -1, 0));
        }

in SdcaBinaryTrainer can produce two different types depending on the specified loss function. To generate model in a type-safe manner, we need two trainers for the two possible model types. cc @TomFinley for visibility.