Hugging Face Open-Sourced FineVision: A New Multimodal Dataset with 24 Million Samples for Training Vision-Language Models (VLMs)