Improving power for robust trans-ethnic meta-analysis of rare and low-frequency variants with a partitioning approach