Using residual regressions to quantify and map signal leakage in genomic prediction