Apple Researchers Introduce FastVLM: Achieving State-of-the-Art Resolution-Latency-Accuracy Trade-off in Vision Language Models