Full boosting loop: assemble a minimal booster

Implement:

def train_gradient_boosting(input, targets, n_trees, max_depth, learning_rate) -> tuple[float, list[dict]]:
    """Returns (initial_prediction, trees)."""

def predict_gradient_boosting(initial_prediction, trees, learning_rate, input) -> np.ndarray:
    """Replays training's accumulation on new data."""

• Reuse 03-gradient-boosting-negative-gradient's fit_tree_to_negative_gradient and 05-regression-trees's predict_regression_tree, don't reimplement either.
• learning_rate scales every tree's contribution, applied identically during training and prediction.