Decision Boundary / Thresholding

Implement:

def predict_labels(p: np.ndarray, threshold: float = 0.5) -> np.ndarray:
    """
    Returns integer labels (0 or 1), same shape as p.
    """