Optimising APY core subsets for scalable genomic prediction