Add macro-averaged Mean Squared Error metric with tests

[nithish-74]

This PR introduces a new regression metric: macro-averaged Mean Squared Error (Macro-MSE).

🔹 What it does

Adds macro_mean_squared_error to imblearn.metrics._regression.py.

Implements computation of MSE per class and returns the unweighted (macro) average.

Includes unit tests in imblearn/tests/test_regression.py to verify correctness.

🔹 Motivation
While sklearn.metrics.mean_squared_error provides sample-wise MSE, there is no built-in option for macro averaging across classes. This is particularly useful in imbalanced regression scenarios where class distributions are skewed, ensuring that minority classes contribute equally to the error metric.

🔹 Changes made

Implemented macro_mean_squared_error function.

Added corresponding test cases with multiple class distributions.

Ensured compatibility with scikit-learn’s API style.

🔹 Next steps

Documentation update (if maintainers prefer exposing this in the public API).

Feedback welcome on naming, placement, and test coverage.