Implement continuous training by allowing an (optional) input with a previously existing model

[jfomhover]

In LightGBM training, you can use a pre-existing model. This does not exist yet in our implementation. We could introduce an optional input for continuous training.

[felixnext]

Likely will also be impacted by the continue train support for the transform-library: microsoft/lightgbm-transform#18