Salesforce AI Research Propose Programmatic VLM Evaluation (PROVE): A New Benchmarking Paradigm for Evaluating VLM Responses to Open-Ended Queries