feat: Add PaddleOCRVL PDF parser with restoration and split table merging

[AdemBoukhris457]

✨ New Features

• PaddleOCRVL PDF Parser: Added new PaddleOCRVLPDFParser class for end-to-end document parsing

  • Uses PaddleOCRVL Vision-Language Model for comprehensive document understanding
  • Single-pass extraction of all content types (headers, text, tables, charts, footnotes, figure titles)
  • Automatic chart recognition and conversion to structured table format
  • Support for pipe-delimited chart content parsing

• Integrated Document Restoration: Optional DocRes image restoration support

  • Configurable restoration tasks (appearance, dewarping, deshadowing, deblurring, binarization, end2end)
  • GPU acceleration support for restoration processing
  • Enhanced page images saved to enhanced_pages/ directory

• Split Table Merging: Automatic detection and merging of tables split across pages

  • Two-phase detection algorithm (proximity detection + structural validation)
  • Configurable thresholds for position, gap, and alignment tolerance
  • Merged tables processed with PaddleOCRVL for complete extraction
  • Visual merge results saved to tables/ directory

• Structured Output Generation:

  • Markdown output with all extracted content
  • HTML output with formatted tables and charts
  • Excel file (tables.xlsx) containing all tables and charts as structured data
  • HTML file (tables.html) with structured tables and charts

🔧 Technical Improvements

• Added platform-specific safetensors dependency handling (Linux/Windows)
• Integrated PaddleOCRVL with existing Doctra architecture
• Robust handling of both dictionary and string-formatted PaddleOCRVL output
• HTML table parsing with BeautifulSoup fallback to regex-based parsing
• Chart content parsing from pipe-delimited format to structured tables
• Progress bars for both notebook and terminal environments

📚 Documentation

• Added comprehensive user guide for PaddleOCRVLPDFParser
• Updated README with parser overview, features, and usage examples
• Added API reference documentation
• Updated core concepts guide with parser description
• Added usage examples in basic-usage.md
• Updated documentation navigation and index

📦 Dependencies

• PaddleOCR with doc-parser support (paddleocr[doc-parser]>=3.2.0)
• PaddlePaddle GPU (paddlepaddle-gpu>=3.2.1) for CUDA 12.6
• Platform-specific safetensors wheels:
  • Linux: safetensors-0.6.2.dev0-cp38-abi3-linux_x86_64.whl
  • Windows: safetensors-0.6.2.dev0-cp38-abi3-win_amd64.whl
• Updated installation documentation with platform-specific instructions

🔄 Migration Notes

• No breaking changes: New parser is additive
• Existing parsers remain unchanged
• PaddleOCR dependencies are optional (parser raises ImportError if not available)
• Users can opt-in by installing PaddleOCR dependencies

📝 Files Changed

Python Code:

• doctra/parsers/paddleocr_vl_parser.py - Main parser implementation
• doctra/parsers/__init__.py - Export new parser
• doctra/__init__.py - Package-level export

Documentation:

• README.md - Added parser documentation and examples
• docs/user-guide/parsers/paddleocr-vl-parser.md - Comprehensive user guide
• docs/api/parsers.md - API reference
• docs/index.md - Updated parser table
• docs/user-guide/core-concepts.md - Added parser description
• docs/examples/basic-usage.md - Added usage example
• mkdocs.yml - Updated navigation

Installation:

• requirements.txt - Added PaddleOCR and safetensors dependencies
• setup.py - Updated with PaddleOCR dependencies
• pyproject.toml - Updated with PaddleOCR dependencies
• docs/getting-started/installation.md - Added installation instructions

✅ Checklist

• Tests pass (parser implementation complete)
• Code formatted and linted
• Documentation updated
• Type hints included
• Docstrings follow project format
• No breaking changes
• Dependencies properly declared