Appends prediction columns to transform schema

[Marcus-Rosti]

Pipelines fail when stages depend on the output of transformSchema since IsolationForest does not append them.

[jverbus]

Thank you for making this suggestion! I already had these checks in the Model class transformSchema method, but you're right that the Estimator should have these too. By adding the new columns at the Estimator stage, downstream pipeline components can anticipate these columns right from the start. This matches the Spark ML design philosophy, where transformSchema at the Estimator level should describe the schema that the resulting Model will produce once fitted.