Accelerated Bayesian Optimal Experimental Design via Conditional Density Estimation and Informative Data