Genetic instruments with too many strings: acknowledging pleiotropy and population structure in Mendelian randomization studies