Exercise M4.04 demo of `LinearRegression` on repeated features is unstable

[ArturoAmorQ]

Originally posted here in the forum. Ridge regression with alpha=1.0 leads to same result as the ordinary linear regression depending on the machine. This is possibly an Issue worth looking at from scikit-learn, but we may alternatively avoid this demo on the MOOC.

Reproducer from OP here: https://gist.github.com/qdpham/21b30181a8e0dd50cfcbba57b61e9b70

[ogrisel]

The problem is not with Ridge which always behaves as expected on both machines but rather that LinearRegression does not always fail as we expect it to fail depending on the machines.