Feature munging pipeline for AutoML

[sebhrusen]

As AutoML wants to apply transformations to the features from the original dataset, we need to encapsulate those transformations in a Pipeline model that can then be returned to the user to ensure that those same transformations are also applied for predictions.

As a parent issue, this is divided into the following tasks:

• AutoML Pipeline – Java API #15855
• AutoML Pipeline – Hyperparameter Search #15856
• AutoML Pipeline – Integration to AutoML #15857
• AutoML Pipeline – Client pipeline description #15859

Independent issues raised for proper implementation of pipeline tasks:

• H2O Core: Consistent mechanism to protect frames or some of their vecs from autodeletion #15858

Optional/future improvements

• Mojo support.
• Provide client API to use those pipelines with any kind of H2O-3 model.