Integrates malicious weight sending with get_model_weights

[joyce-yuan]

Integrated malicious weight sending with get_model_weights by introducing new parameter get_external_repr: bool.

    def get_model_weights(self, get_external_repr:bool=True) -> Dict[str, Tensor]:
        """
        Share the model weights

        Args:
            get_external_repr (bool): Whether to get the external representation of the model, 
            used for malicious attacks where the model weights are modified before sharing.
        """

If we are trying to get the external representation of a malicious node, it'll call get_malicious_model_weights. If internal representation, it'll just return model weight (i.e. for training)