Dissecting gene expression heterogeneity: generalized Pearson correlation squares and the K-lines clustering algorithm