Addressing the Length Bias Problem in Document-Level Neural Machine Translation