Improve PDF parsing

[cbldev]

Actual behavior

I noticed that on some complex PDF, with tables, pdftotext produce better result than pdf-reader gem.

pdftotext: https://www.xpdfreader.com/pdftotext-man.html

Issue in Langchainrb: patterns-ai-core/langchainrb#682

Expected behavior

Good results on complex PDF parsing.

[cbldev]

Docling seems to be a better option

[cbldev]

I started a documentation with Docling Serve: https://github.com/dilolabs/nosia?tab=readme-ov-file#with-docling-serve

Need to update the Docling Serve Docker Compose file to adapt it to different GPUs: https://github.com/dilolabs/nosia/blob/main/docker-compose-docling.yml