An advanced open-source multimodal large language model from the InternVL3.5 family, specializing in versatile vision-language tasks, and enhanced reasoning