Add interpretation of vanishing R² score in terms of DummyRegressor

[ArturoAmorQ]

In the Comparing model performance with a simple baseline notebook we introduce the DummyRegressor as baseline using MAE.

I think it's an opportunity to mention that, from the mathematical definition of the R² score, a model that always predicts the mean results in a zero training R² score. Therefore, the strategy="mean" of the sklearn.dummy.DummyRegressor can be considered an implicit baseline for such score.

Having this explanation would help interpreting the scores in the Non i.i.d data notebook, then I consider this issue to be part of #786.