Support GANS

[MaxVanDijck]

Due to the way we currently initialize and load models:

# Create Model and Load model if applicable
    model = idist.auto_model(hydra.utils.instantiate(config.model))
    log.info(f"Model: {model}")

    if config.checkpoint:
        model.load_state_dict(torch.load(config.checkpoint), strict=False)

It makes the refactoring of src/train_pipeline.py necessary if users want to use multiple multiple models i.e in the use of GANs.
Ideally this will be changed to support both single + multi model purposes.
Creating this issue as a reminder and to put ideas.