Add support for transform_fn

At the moment, you can only provide 'input_fn', 'predict_fn' and 'output_fn' which limits flexibility. I have a use case that requires accept_type to be used by the predict_fn, which is not possible in this structure. MXNet containers have the option to use 'transform_fn'.

I think this line limits the container to 'input_fn', 'predict_fn' and 'output_fn' :
https://github.com/aws/sagemaker-scikit-learn-container/blob/master/src/sagemaker_sklearn_container/serving.py#L90