Support for incremental training on a schedule, i.e. take the last output model and start new training run

[cosmicBboy]

As an ML engineer I want to schedule a training job to run on some cadence and support starting from the last invocation of the training workflow.