Cleanup ElasticNet, Lasso, and CD

[jcrist]

This is a general python/cython cleanup of our coordinate-descent based estimators (and the solver itself).

• Updates the code to follow modern cuml and sklearn conventions
• Moved all hyperparameter validation out of __init__ and into respective fits, keeping the inits simple (as per sklearn convention). Also improved hyperparameter validation a bit.
• Improved data validation, including checking for minimal data input sizes. This leads to a few upstream sklearn tests passing.
• Split out the fit from CD into a top-level function, as described in Deprecate and remove solver_model #6938. I didn't touch QN in this PR, that will be done in a followup.
• Removed the solver_model attribute on ElasticNet/Lasso. Since this attribute was undocumented and (mostly) unaccessed in our tests, I view it as an implementation detail and not something that necessarily needs a deprecation period.
• Cleaned up docstrings a bit
• General code tidyness. Hopefully this is a bit more readable now.

Part of #7317.
Part of #6938.

[jcrist]

• Split out the fit from CD into a top-level function, as described in Deprecate and remove solver_model #6938. I didn't touch QN in this PR, that will be done in a followup.
• Removed the solver_model attribute on ElasticNet/Lasso. Since this attribute was undocumented and (mostly) unaccessed in our tests, I view it as an implementation detail and not something that necessarily needs a deprecation period.

In the long run I hope we can deprecate our solver estimators as public interfaces (sklearn doesn't expose their solvers directly, why should we?), instead relying on pure functions for fitting and our generic linear model predict/decision_function implementations for prediction. There's no need to have a CD or QN specific predict/decision function, these are solvers for fitting alone, which should let us drop some C++ code and simplify the wrappers. That can all be done in followups though, for now the focus here was on tidying up ElasticNet and Lasso.

[viclafargue]

LGTM!

[jcrist]

/merge