Describe the workflow you want to enable

Hi @ scikit-learn devs!

Over at pyfixest, we have implemented a Frisch-Newton Interior Point solver to fit quantile regressions. The algorithm goes back to work from Koenker. In practice, we have followed Koenker and Ng "A Frisch-Newton Algorithm for Sparse Quantile Regression".

The code is licensed under MIT and available here.

We (@apoorvalal) have collected some benchmarks here - the FN solver seems to outperform the scikit default solver by an order of a magnitude.

Would you be interested in a PR that adds the FN solver as a new estimation method to the quantile regression class?

We've also implemented algorithms from Chernozhukov et al that can drastically speed up estimation of the entire quantile regression process.

All the best, Alex

Describe your proposed solution

I open a PR and add a new solver "fn" to QuantileRegressor.

Describe alternatives you've considered, if relevant

No response

Additional context

No response