Text this: Assessing non-inferiority for binary matched-pairs data with missing values: a powerful and flexible GEE approach based on the risk difference