Binary learners need to have IComponentFactory implementation in order to work with OVA.

[Ivanidzo4ka]

With #622 OVA expects predictor type to be IComponentFactory, and in current moment only learner which exposed itself as component is FastTree, and it exposes itself only as ITrainer instead of ITrainer<IPredictorProducing> required for OVA.

Which mean we have to go through all our Binary learners and expose them as ComponentKind and create proper factories. Same probably states for Regression, MultiClass, and Ranking.