Add train_deviance_path_ to GeneralizedLinearRegressorCV

[Matthias Schmidtblaicher (MatthiasSchmidtblaicherQC)]

Addresses last point on #933 .

GeneralizedLinearRegressorCV now exposes train_deviance_path_ alongside the existing deviance_path_ (validation deviance). This makes it easy to compare training and validation performance across the regularization path.

• New attribute train_deviance_path_ with shape (n_folds, n_l1_ratios, n_alphas), matching deviance_path_.
• Computed inside the existing CV loop with negligible overhead (reuses the already-unstandardized coefficients).
• Saves a reference to x_train before standardization so train deviance is evaluated on the original scale.

Checklist

• Added a CHANGELOG.rst entry

[Copilot]

Pull request overview

This PR extends GeneralizedLinearRegressorCV to expose training-set deviance values along the CV regularization path, enabling direct comparison between training and validation performance across folds, l1_ratios, and alphas.

Changes:

• Add train_deviance_path_ computed within the existing CV loop in GeneralizedLinearRegressorCV.
• Add a new test validating the presence and shape of train_deviance_path_.
• Add a CHANGELOG.rst entry documenting the new attribute.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
src/glum/_glm_cv.py	Computes and stores train_deviance_path_ alongside existing deviance_path_ and updates attribute docs.
tests/glm/test_glm_cv.py	Extends solver-equivalence assertions and adds a new test for train_deviance_path_.
CHANGELOG.rst	Documents the new train_deviance_path_ feature for the next release.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.