Pass two-dimensional endpoints to plot_linear_regression estimators

[ktz03]

Code of Conduct

Reviewed the repository's Code of Conduct.

Description

plot_linear_regression fits an estimator using a two-dimensional feature array, but passes scalar endpoints to predict. The default scikit-learn LinearRegression therefore rejects even the existing User Guide example with Expected 2D array, got scalar array instead.

Pass each endpoint as a single-feature sample and extract its prediction. Keep the fitted coefficients, correlation, line limits and plotting options unchanged. Add regressions for the documented data, list/array/column inputs, explicit and automatic limits, LinearRegression/Ridge, and custom correlation/legend options.

Related issues or pull requests

Fixes #1252

Pull Request Checklist

• Added a note to ./docs/sources/CHANGELOG.md.
• Added regression tests in ./mlxtend/plotting/tests/test_plot_linear_regression.py.
• Modified the corresponding User Guide notebook: not applicable; its existing example is restored without an API change.
• Ran the focused test file with branch coverage: 14 passed. The same final tests on the unmodified implementation report 14 failed. This is the focused verification allowed for small changes by the template; the full repository suite was not run.
• Checked style: changed-file Black and isort passed. Repository-wide flake8 reports the same pre-existing W293 in frequent_patterns/tests/test_association_rules.py:62 on baseline and fixed code.

The broader plotting test directory cannot collect on either baseline or fixed code because its existing test_pca_corr_graph.py compares a Version object with a string. No test or CI threshold was changed.

Validated locally on Windows, Python 3.12.14, NumPy 2.4.0, SciPy 1.16.3, scikit-learn 1.8.0 and Matplotlib 3.10.8 (Agg backend). No other platform, interactive display or upstream CI pass is claimed.