Min-Max Scaler

Scale every feature into [0, 1] using statistics learned on the training data only.

Implement a class MinMaxScaler with:

• fit(X): store each column's min and max, return self
• transform(X): (X - min) / (max - min). A constant column (max == min) maps to 0.
• inverse_transform(Z): undo the scaling

Test data can fall outside the training range, so transformed values may be below 0 or above 1. That's expected.