Feature importance from a fitted tree

Implement:

def feature_importances(
    tree: dict, input: np.ndarray, labels: np.ndarray, n_features: int
) -> np.ndarray:
    """
    tree: a tree from build_tree (03-best-split-minimal-tree), fitted
    on (input, labels).

    Returns:
        shape (n_features,), summing to 1 (or all zeros for a
        single-leaf tree).
    """

• Reuse 02-information-gain's information_gain, don't reimplement it.
• Re-derive each node's data subset by walking the tree with input/labels, the same way predict_tree walks it, rather than assuming the tree stores its own data.