refactor VLM configuration to use dependency pattern

[AdemBoukhris457]

Summary

This MR refactors VLM (Vision Language Model) configuration across all parser classes to use a dependency pattern, similar to the previous OCR engine refactoring. Instead of exposing individual VLM parameters (use_vlm, vlm_provider, vlm_model, vlm_api_key) directly in parser constructors, users now initialize a VLMStructuredExtractor instance externally and pass it to the parser.

Changes

Code Changes

• Parser Classes: Refactored StructuredPDFParser, EnhancedPDFParser, ChartTablePDFParser, and StructuredDOCXParser to accept an optional vlm: Optional[VLMStructuredExtractor] parameter
• CLI: Updated all CLI commands (parse, enhance, parse-docx, extract) to create VLMStructuredExtractor instances before passing to parsers
• UI Components: Updated all Gradio apps (gradio_app.py, hf_space/app.py, hf_space/app_fixed.py, doctra/ui/full_parse_ui.py, doctra/ui/enhanced_parser_ui.py) to create VLM engine instances based on user inputs

Documentation Changes

• README.md: Updated all VLM usage examples and added VLM Engine Configuration section
• Documentation: Updated all docs files including:
  • VLM integration guide
  • API reference
  • Parser guides
  • Usage examples
  • Quick start guide

Benefits

1. Clearer API: VLM configuration is now separated from parser logic, making the API more intuitive
2. Prevents Mixed Configurations: Users can't accidentally mix VLM settings, as they must create a single, pre-configured engine
3. Engine Reuse: VLM engines can be created once and reused across multiple parsers
4. Consistency: Matches the pattern already established for OCR engines
5. Better Error Handling: VLM initialization errors are caught earlier, before parser creation

Migration Guide

Before:

parser = StructuredPDFParser(
    use_vlm=True,
    vlm_provider="openai",
    vlm_api_key="your-key"
)

After:

from doctra.engines.vlm.service import VLMStructuredExtractor

vlm_engine = VLMStructuredExtractor(
    vlm_provider="openai",
    api_key="your-key"
)

parser = StructuredPDFParser(vlm=vlm_engine)