fix: constrain numpy<2.4 and document format_block_content image path behavior

[TingquanGao]

Summary

• Fix PaddleOCR-VL同样的代码能运行在Ubuntu 25.04/Python 3.13.3，但是Ubuntu 24.04/Python 3.12.3出错 PaddleOCR#17495: Constrain numpy version to `>=1.24, <2.4` in `setup.py`. numpy 2.4 promotes the implicit non-0-d array to Python scalar conversion from `DeprecationWarning` to `TypeError`, causing `RuntimeError` in PaddleX. Adding the upper bound prevents incompatibility.
• Fix PaddleOCR-VL解析json数据中图片block没有图片地址 PaddleOCR#17143: Supplement documentation for `format_block_content` parameter in PaddleOCR-VL and PP-StructureV3 pipeline docs (both Chinese and English) to clarify the image path behavior:
  • When `format_block_content=True`: image-type block's `block_content` includes image path info (e.g. `<img src="..." />`)
  • When `format_block_content=False` (default): image-type block's `block_content` only contains OCR-recognized text, no image path
  • To get image paths in JSON output, set `format_block_content=True`

Changes

• `setup.py`: Add numpy upper bound `< 2.4`
• `docs/pipeline_usage/tutorials/ocr_pipelines/PaddleOCR-VL.md`: Supplement `format_block_content` description
• `docs/pipeline_usage/tutorials/ocr_pipelines/PaddleOCR-VL.en.md`: Supplement `format_block_content` description
• `docs/pipeline_usage/tutorials/ocr_pipelines/PP-StructureV3.md`: Supplement `format_block_content` description
• `docs/pipeline_usage/tutorials/ocr_pipelines/PP-StructureV3.en.md`: Supplement `format_block_content` description

Test plan

• Verify `numpy>=2.4` is no longer installed when following PaddleX install instructions
• Verify JSON output contains image paths when `format_block_content=True`
• Verify JSON output does NOT contain image paths when `format_block_content=False`

Fixes PaddlePaddle/PaddleOCR#17495
Fixes PaddlePaddle/PaddleOCR#17143

🤖 Generated with Claude Code

[paddle-bot]

Thanks for your contribution!