enh: translate the coeficients of linear model to real scale

[MarieSacksick]

Is your feature request related to a problem? Please describe.

Due to the scaling done prior to linear models, we can not interpret the coefficient values with regards to the original unit of the feature, while it is interesting, in particular in regression.
It would be nice to provide easily unscaled coefficients, without forgetting the intercerpt, so that the coefficients can interpreted using the original units.

Describe the solution you'd like

Maybe what is done in

skore/examples/model_evaluation/plot_feature_importance.py

Lines 259 to 265 in a577cad

	# %%
	import numpy as np
	unscaling_weights = np.insert(ridge_report.estimator_[0].scale_, 0, 1, axis=0).reshape(
	-1, 1
	)
	ridge_report.feature_importance.coefficients() * unscaling_weights

(lines 259 to 265)

Describe alternatives you've considered, if relevant

No response

Additional context

cf the example plot_feature_importance.py

[glemaitre]

It could be tricky when you have a complex preprocessing, e.g. a step with PolynomialFeatures or PCA. So we would need to think thoroughly this feature to know in which case we can provide it and thus what should control the behaviour.

[MarieSacksick]

Yes, at first I thought it could be a "good first issue" since the code was available in Sylvain's notebook. And then I realized it wouldn't be :).