Confronting false discoveries in single-cell differential expression