Implement the independent standard error as a metric

[mastoffel]

The ISE is a metric for models where the predict method has a return_std argument. It essentially calculates how many of the true values fall within 2 std of the predictions, so something like this:

def ise(estimator, X, y_true):
    """
    Scorer: the Independent Standard Error (ISE).
    
    Parameters:
    y_true (array-like): True output values
    y_pred (array-like): Predictions from the model (should be a tuple of mean and std)
    
    Returns:
    float: ISE value as a percentage
    """
    if not has_predict_with_std(estimator):
        return np.nan
    y_pred_mean, y_pred_std = estimator.predict(X, return_std=True)
    within_2std = np.abs(y_pred_mean - y_true) < 2 * y_pred_std
    ise_value = np.mean(within_2std) * 100  
    return ise_value

The problem is that this can't just be wrapped using make_scorer in cross_validate.py, because it requires predict to be called with return_std=True (default is False). And then it needs to work with multioutput too.