Information Gain for a split

Implement:

def information_gain(
    parent_labels: np.ndarray,
    left_labels: np.ndarray,
    right_labels: np.ndarray,
) -> float:
    """
    parent_labels: labels of every sample in the node before splitting.
    left_labels, right_labels: parent_labels partitioned by the
    candidate split, every sample in exactly one of the two.

    Returns:
        how much the split reduces Gini impurity, a float.
    """

• left_labels and right_labels together contain every sample in parent_labels, in some order, no sample is dropped or duplicated.
• Reuse 01-gini-impurity's gini_impurity rather than reimplementing it.