@browserbasehq/stagehand patch Fix agent.execute() occasionally reporting a completed run as failed (most often with reasoning models such as openai/gpt-5.x).