Update counterfactual.py

[mariam851]

This pull request adds an optional parameter feature_names_to_vary to the
create_counterfactual function in mlxtend.

With this feature, users can freeze certain features while generating
counterfactual explanations, allowing only selected features to vary during
optimization. This is useful for cases where some features are immutable
or should remain constant for practical or ethical reasons.

Changes include:

• Added feature_names_to_vary parameter (list of indices or boolean mask)
• Updated distance calculation and loss function to respect frozen features
• Added warnings for unsupported inputs (e.g., string feature names)
• Maintained backward compatibility for existing code

Tested locally with a small example to ensure the function respects frozen
features and outputs valid counterfactuals.

This is an experimental feature and feedback is welcome.

[rasbt]

Thanks for the PR. The feature looks good overall, but similar to the other PR, I suggest removing the LLM-generated comments that explain what's being added as new, because outside this PR, it would clutter the code and be redundant.