[FEAT] Add code to return model and featurizer from anvil spec outside of `predict`

[hmacdope]

Currently to do inference with a model, one has to construct and then destruct a model using the inference.predict module. There should be an option to have the model returned to the user to live in memory, as this is much more resource efficient (e.g on HuggingFace)

[hmacdope]

This actually already exists as load_anvil_model_and_metadata here but may need some tweaking