Type I error rates of rare single nucleotide variants are inflated in tests of association with non–normally distributed traits using simple linear regression methods