Improved Policy Evaluation for Randomized Trials of Algorithmic Resource Allocation