What problem does this PR solve?

Docling requires Python 3.12 but the Dockerfile installs 3.11, causing compatibility issues. The docling_parser.py uses outdated API patterns without leveraging modern configuration options for improved accuracy and error handling.

Type of change

• Bug Fix (non-breaking change which fixes an issue)
• New Feature (non-breaking change which adds functionality)
• Documentation Update
• Refactoring
• Performance Improvement
• Other (please describe):

Changes

Dockerfile

• Updated base stage: uv python install 3.11 → 3.12

pyproject.toml

• Added dependency: docling>=2.60.0,<3.0.0

deepdoc/parser/docling_parser.py (276 insertions, 70 deletions)

Modern Docling API integration:

• PdfPipelineOptions with TableFormerMode.ACCURATE for better table extraction
• EasyOcrOptions configuration for OCR
• PyPdfiumDocumentBackend support
• Converter instance caching to avoid repeated initialization

Enhanced error handling:

• Try-catch blocks with detailed logging (exc_info=True)
• Failure state caching to prevent DoS from repeated init attempts
• Graceful degradation when optional features unavailable

Security hardening:

• Path traversal prevention: sanitize filenames before writing temp files
• Safe backend instantiation with exception handling
• Input validation for file paths

Example of improved configuration:

# Before: basic initialization
conv = DocumentConverter()

# After: configured with caching
def _get_converter(self):
    if self._converter is not None:
        return self._converter
    
    pipeline_options = PdfPipelineOptions(
        do_table_structure=True,
        table_structure_options=TableStructureOptions(
            mode=TableFormerMode.ACCURATE,
            do_cell_matching=True
        )
    )
    self._converter = DocumentConverter(format_options={
        InputFormat.PDF: PdfFormatOption(
            pipeline_options=pipeline_options,
            backend=PyPdfiumDocumentBackend()
        )
    })

All changes maintain backward compatibility. Existing DoclingParser() usage unchanged.