[BUG] Remove the hack that re-do the aggregation in fedeca robust mode

[jeandut]

Remove the hack in fedeca_core.py:

                    # Awful but hard to hack better
                    # compute_plan_keys[2] is a tuple with
                    # (scores, train_states, aggregate_states) in it
                    self.variance_matrix = sum(
                        [
                            e.algo._client_statistics["Qk"]
                            for e in self.compute_plan_keys[2][1].state
                        ]
                    )

And replace it with sthg like:

from substrafl.model_loading import download_aggregate_shared_state
self.variance_matrix = download_aggregate_shared_state(
            client=self.ds_client,
            compute_plan_key=compute_plan.key,
            round_idx=0,
        )

To test in simulation mode.