Unsuitability of likelihood-based asymptotic confidence intervals for Response-Adaptive designs in normal homoscedastic trials