Gini Impurity for a split

Implement:

def gini_impurity(labels: np.ndarray) -> float:
    """
    labels: 1-D array of class labels (any integers, not necessarily
    0..k-1 or contiguous), one per sample in a single tree node.

    Returns:
        the Gini impurity of this set of labels, a float in [0, 1).
    """

• Works for any number of classes, not just binary.
• Labels don't need to be 0..k-1, use whatever distinct values are actually present.
• An empty labels array has no impurity by convention, return 0.0.
• A node where every label is identical is perfectly pure, 0.0.