A Multi-Fidelity Approach to Testing and Evaluation of AI-Enabled Systems