On Efficient Estimation of Distributional Treatment Effects under Covariate-Adaptive Randomization