Pin nanonets-ocr2.py + paddleocr-vl.py to stable vLLM (>=0.15.1)

Fix config_name=None breaking push_to_hub + use cu129 nightly index

Update PaddleOCR-VL configuration for vLLM compatibility and enhance model parameters

Add pyarrow to dependencies in paddleocr-vl.py

Refactor dependencies and clean up whitespace in paddleocr-vl.py

Add PaddleOCR-VL script for document processing with vLLM support