Full Training Loop

Implement:

def train_logistic_regression(
    input: np.ndarray,
    target: np.ndarray,
    lr: float,
    epochs: int,
) -> tuple[np.ndarray, np.ndarray]:
    """
    input:  shape (batch_size, in_features)
    target: shape (batch_size,), 0 or 1 per sample

    Returns:
        weight: shape (1, in_features)
        bias: shape (1,)
    """

Reuse 01-hypothesis-function's linear, 01-sigmoid's sigmoid, 03-bce-gradient's bce_gradient, and Linear Regression's 04-gd-step's gd_step, rather than reimplementing any of their logic.