v1.10.0: Table Extraction Benchmarking

🔢 Table Extraction Benchmarking

This release adds comprehensive table extraction benchmarking capabilities with separate workflows to avoid contaminating general text extraction benchmarks.

✨ New Features

• Separate Table Extraction Workflow: Isolated benchmarking with specialized timeouts (2-8 hours)
• GMFT Support for Kreuzberg: Advanced table extraction with kreuzberg[gmft]
• Table Analysis Module: Structure preservation and content detection scoring
• Table-Specific CLI: --table-extraction-only flag for targeted benchmarking
• Extended Test Documents: HTML tables with simple and complex structures

🚀 Framework-Specific Optimizations

• Kreuzberg: GMFT integration for advanced table processing (8h timeout)
• Docling: Built-in ML table extraction (6h timeout)
• Unstructured: Hi-res strategy for table detection (5h timeout)
• MarkItDown: Basic table conversion (3h timeout)
• Extractous: Fast but limited table support (2h timeout)

📊 Analysis Capabilities

• Table structure preservation scoring
• Content detection and accuracy metrics
• Framework performance rankings
• Format support matrix analysis
• Comprehensive reporting (CSV, Markdown, JSON)

🔧 Usage

# Run table extraction benchmarks
uv run python -m src.cli benchmark --table-extraction-only --framework all

# Analyze table extraction results  
uv run python -m src.cli table-analysis --results-dir results

Both general text extraction and table extraction workflows will trigger automatically on this release.