[feat] Add multimodal PDF reader (text + image captions)

[Bessouat40]

Summary

This PR introduces a new VllmPDFReader, a multimodal PDF ingestion component that extracts:

• Text from every page
• Vision captions for each image via a VLLM / OpenAI-compatible model

Previously, image information was silently lost during ingestion, especially problematic for:

• Technical documentation
• Engineering reports
• Training manuals
• Scientific papers including diagrams, schematics, and figures

This new reader ensures full preservation of PDF semantics, producing separate Document objects for both text and image captions.

Related issue

This PR relates to #4677 (similar in goal although not a direct implementation).

Type of change

• Bug fix
• New feature
• Breaking change
• Improvement
• Model update
• Other:

---

Checklist

• Code complies with style guidelines
• Ran format/validation scripts (./scripts/format.sh and ./scripts/validate.sh)
• Self-review completed
• Documentation updated (comments, docstrings)
• Examples and guides: Relevant cookbook examples have been included or updated (if applicable)
• Tested in clean environment
• Tests added/updated (if applicable)