Expected Information Gain Estimation via Density Approximations: Sample Allocation and Dimension Reduction