A remark on normal forms and the "upside-down" $I$-method for periodic NLS: Growth of higher Sobolev norms