Meet dots.ocr: A New 1.7B Vision-Language Model that Achieves SOTA Performance on Multilingual Document Parsing