Evaluating COVID-19 vaccine allocation policies using Bayesian $m$-top exploration