Allow Apeiron to use custom checkpoints

[anagainaru]

Model harnesses can define custom checkpoints. If the model requires metadata beyond weights (e.g., preprocessing scalers, feature names, architecture parameters). This PR introduces an optional build_checkpoint_payload() function that the model harness can overwrite (by default only weights are saved which is the current behavior of Apeiron).

def build_checkpoint_payload(self):
    """Save checkpoint with all metadata needed for loading."""
    return {
        "model_state_dict": self.model.state_dict(),
        "feature_names": self._feature_names,
        "scaler": self._scaler,
        "input_dim": self._input_dim,
    }

This will ensure that the checkpoints and the input model can have the same elements (i.e. the checkpoints can be used as input model for Apeiron using the same model harness used to create them)

Documentation

• Docstrings updated
• User docs / README updated

Testing

• Testing on the AERIS model harness that requires a feature_list in addition to weights

Closes #125