ENH – Add skglm SmoothQuantileRegressor solver and related improvements (add. python based solvers, sparse dataset (finance), lambda_max fix)

[floriankozikowski]

Context of the PR

skglm.experimental.SmoothQuantileRegressor was first introduced in sklearn-contrib/skglm #312.
This PR adds this to this benchopt benchmark.

While wiring it into benchmark_quantile_regression, I noticed a few gaps that prevented fair comparisons and reproducible results, so I addressed them in the same pass (sparse dataset (finance), lambda_max).
For broader comparison, I also added various other python based solvers.

Contributions of the PR

• Adds skglm.experimental.SmoothQuantileRegressor
• Adds sklearn.linear_model.QuantileRegressor
• Adds asgl.Regressor
• Adds statsmodels.QuantReg
• Introduces a sparse finance dataset (https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/regression.html#E2006-tfidf)
• Fixes lambda_max computation in the objective function for sparse data
• Add config.yml for easier solver configuration runs

[floriankozikowski]

Hi @Badr-MOUFAD. I wanted to do some benchmarking for the new skglm QuantileHuber solver. In this run, I also added a few other solvers & reviewed the lambda max computation in the objective.py, which I think is more accurate now.
As skglm really excels with sparse data, I added the finance dataset where no other solver really worked for a more broad comparison.

I am happy for review. If it's too much in one PR let me know and I can delete stuff again. I did the benchmark for myself and thought it might be useful to push here.

P.S.: I also made a script for an R solver for quantile regression (as R is more widely used for these things), if you want I can add this also to the repo (either in this PR or probably better in a new seperate PR).