fix: robust page number extraction across pymupdf4llm versions

feat: page-level chunking with page_number metadata + vision_descriptions support

fix: make chonkie internal imports optional to avoid startup crash

Upload utils.py with huggingface_hub