Add SaveEnsembleMixin

[Mr-Geekman]

🚀 Feature Request

We want to make ensembles saveable.

Proposal

1. Add class SaveEnsembleMixin(SaveMixin), it will be responsible for saving all ensembles.
2. Signature of save for this moment remains the same. But in the future we plan to make it different.

   def save(self, path: pathlib.Path, save_ts: bool = False).

   It will make it possible to save ts with pipeline. For now, we don't want to implement saving TSDataset (it will be much simplier after completion of TSDataset-v2.0).
3. Signature of load we want to change.

   def load(cls, path: pathlib.Path, ts: Optional[TSDataset] = None) -> Any.

   We expect that user gives us dataset we want to work with. If user hasn't give us the dataset, may be he wants to refit pipeline or he want to give a custom TSDataset into forecast/predict.
4. How save should work.
   • Remove attributes pipelines, ts.
   • Save object with SaveMixin.
   • Save each pipeline in pipelines separately into folder inside the archive.
5. How load should work.
   • Load object with SaveMixin.
   • Load pipelines. We should get classname from meta or implement load method that makes this logic with loading classname for us.

Test cases

1. Test mixin.
   • Test in saving that everything except for pipeline is saved in object.pkl.
   • Test in saving that pipelines are saved into separate folder and can be loaded from it.
   • Test error on non-existent file.
   • Test that load restores pipelines and ts (if given).
   • Test warning on different versions on etna.
2. Test saving/loading on VotingEnsemble.
3. Test saving/loading on StackingEnsemble.
4. Test saving/loading on DirectEnsemble.

Additional context

No response

[Mr-Geekman]

Closed via #1046.

[Mr-Geekman]

Closed via #1046.