How does embedding correction with Harmony affect the reduction of technical batch effects in snATAC-seq datasets?