Design and quality control of large-scale two-sample Mendelian randomisation studies