Warn when encoders are ignored by fit/predict_from_dataset

[webzuweb]

Fixes #3006.

Summary

add_encoders is only applied when training/predicting via fit()/predict(). When using fit_from_dataset()/predict_from_dataset(), encoders are silently ignored, which is a blind spot for users.

Per the discussion in #3006, this PR adds a logger.warning whenever fit_from_dataset()/predict_from_dataset() is called on a model initialized with add_encoders, and documents the behavior in both docstrings.

Changes

• TorchForecastingModel.fit_from_dataset(): warn + docstring note when self.add_encoders is set.
• TorchForecastingModel.predict_from_dataset(): warn + docstring note when self.add_encoders is set.
• Tests: test_encoders_ignored_in_from_dataset_warns and test_encoders_not_set_from_dataset_no_warn.

No behavior change for models without encoders.

[dennisbader]

Hi @webzuweb and thanks for the PR. Closing it however since it's a duplicate of #3086.