Random Forest: majority vote aggregation

Implement:

def random_forest_predict(trees: list[dict], input: np.ndarray) -> np.ndarray:
    """
    trees: a list of already-built trees.
    Returns: shape (n_samples,), majority vote across all trees.
    """

• Reuse 03-best-split-minimal-tree's predict_tree, don't reimplement tree traversal here.
• Ties are broken by the lower class label (a natural side effect of using sorted class values, no special-case code needed).