[DEE-6] [FEAT] Better error message when passing illegal columns to _DummyModel

[noamzbr]

Is your feature request related to a problem? Please describe.
As raised by @OlivierBinette in #2202, sklearn models can run correctly when supplied with additional columns (on top of the feature columns), while _DummyModel must receive exactly the feature columns. This may lead to confusion when implementing custom checks, as users may expect context.model.predict(dataset.data) to work.

Describe the solution you'd like
In these cases, the _DummyModel should raise an error explaining it expects to receive exactly the feature columns, and no more.

_DEE-6