[MRG] ENH add macro-averaged mean squared error

[Ab2nour]

Reference Issue

Closes #846

What does this implement/fix? Explain your changes.

New metric for ordinal classification: macro_averaged_mean_squared_error, following the same logic as the previously introduced macro_averaged_mean_absolute_error.

Changes:

• implemented macro_averaged_mean_squared_error
• added new metric entry, reference and link in the documentation
• exported function in __init__.py
• added minimal test case & registered test public function
• CI passing (preview in my fork)

Any other comments?

A comment in the issue suggested adding a squared=True parameter.

As of scikit-learn 1.4 (see scikit-learn/scikit-learn#26734), this is now outdated: the new function is standalone.

Let me know if further changes are needed.