Estimating prevalence of human traits among populations from polygenic risk scores