If scale_pheno is TRUE, the phenotypes are
mean-centered and standardized before running
the svd.
Because we use SVD in this step, there can be
no missing values in the phenotype matrix. Any individuals
with missing values are removed with a message.