The environment–growth dilemma: new evidence using a panel cointegration approach