Gravar-mail: Efficient estimation of ideal-observer performance in classification tasks involving high-dimensional complex backgrounds