On multivariate deconvolution with Wasserstein loss: minimax rates and Bayesian contraction rates