Add surrogate training

[Crivella]

Changes

• Add training logic for the surrogate model
  • Include logic for transfer-learning (start from pre-trained model and run with frozen layers)
  • Include logic for cross-validation learning (uses a KFold split of a single set of data to automatically generate the train/validation/test subsets)
• Add wood_ms train CLI command to start a surrogate training

Refactor

• Split pipeline logic into a separate class