Optimal configurations for imaging polarimeters: impact of image noise and systematic errors