We need to be able to keep Stagehand open to the latest and greatest. If Stagehand is the web agent SDK, we need to allow users to plug-and-play new vision and language models with ease. We should be able to support just-vision, or even an ensemble of models as well.