feat: skore.cross_validate should support matplotlib, altair (and plotly)

[sylvaincom]

This is priority P2.

Problem

skore's cross_validate displays the results using plotly.

But there are issues with plotly. We found that plotly does not render well in the Sphinx documentation (#594) so we opened issues to plotly and to sphinx. Indeed, this plotly rendering issue appears also in the official Sphinx doc.

Related: plotly/Kaleido#209

Nevertheless, while @augustebaum first started using altair, with @MarieS-WiMLDS we recommended a switch from altair to plotly as we believe data scientists in the industry are more familiar with plotly. @koaning prefers altair over plotly. Note that, apparently, the plotly rendering issue on Sphinx also applies to altair.

Suggestion

Let's support both plotly and altair (and matplotlib) and have plot_backend argument for the users to specify which one they prefer?

Something like:

from skore import cross_validate
...
cv_results = cross_validate(Ridge(), X, y, cv=5, project=my_project, plot_backend="altair")

Example: pandas has a plot method that does matplotlib by default but you can also have plotly:
https://plotly.com/python/pandas-backend/

[tuscland]

It is difficult to support several backends for every assistive feature.
If the problem is cosmetic, users can however access the underlying data and draw their own plot using the library of their choice.