feat: Add PaddleOCR-VL document converter

[Bobholamovic]

Related Issues

Proposed Changes:

This is a new feature that adds the official PaddleOCR integration for Haystack, providing a PaddleOCR-VL document converter component. The component leverages PaddleOCR's API for document parsing and supports text extraction from both PDF and image files.

How did you test it?

This PR includes a complete unit test suite including initialization tests, parameter validation, file type inference, API call tests, etc. Tests cover PDF and image file processing.

Notes for the reviewer

Checklist

• I have read the contributors guidelines and the code of conduct
• I have updated the related issue with new insights and changes
• I added unit tests and updated the docstrings
• I've used one of the conventional commit types for my PR title: fix:, feat:, build:, chore:, ci:, docs:, style:, refactor:, perf:, test:.

[CLAassistant]

All committers have signed the CLA.

[anakin87]

Hey... thanks for the implementation!

I created #2569 to track the work to be done.

I don't think that I will be able to review this PR in detail soon, but in the meantime, please add a CI workflow similar to https://github.com/deepset-ai/haystack-core-integrations/blob/main/.github/workflows/anthropic.yml to make sure that tests run in the CI.

[Bobholamovic]

Hey... thanks for the implementation!

I created #2569 to track the work to be done.

I don't think that I will be able to review this PR in detail soon, but in the meantime, please add a CI workflow similar to https://github.com/deepset-ai/haystack-core-integrations/blob/main/.github/workflows/anthropic.yml to make sure that tests run in the CI.

Thanks. The CI workflow has been added.

[anakin87]

Thanks for the implementation.

I did a first pass and found some opportunities for improvement...

[anakin87]

Could you explain why PaddleOCRVLDocumentConverter instead of a more generic PaddleOCRDocumentConverter? Is this component tied to a specific model? It might be used with other models?

[Bobholamovic]

PaddleOCR offers two document-parsing solutions: PP-StructureV3 and PaddleOCR-VL, and the interfaces are slightly different. The PaddleOCRVLDocumentConverter uses the PaddleOCR-VL algorithm internally, which is why it is named this way. We planned to add a PPStructureV3DocumentConverter in a future PR. Would you recommend using a single all-in-one class that supports the different PaddleOCR algorithms as backends, or separate classes as we originally planned?

[anakin87]

I don't know the differences in terms of interface. I just would not like to have almost identical components but I trust you on this aspect. Please provide more information if possible.

[Bobholamovic]

If we define two separate converters, I expect the following differences between them:

1. The optional parameters accepted by their __init__ methods differ significantly. You can see the differences in the supported arguments for the two algorithms here and here.
2. Although the run method of each converter outputs raw_paddleocr_responses with a broadly similar structure, the exact fields are not identical, and there are notable differences.