Months before OpenAI’s artificial intelligence went rogue and attacked the A.I. start-up Hugging Face, two employees raised an alarm with top executives that its newest A.I. models were not being appropriately monitored during testing. They were ignored.