fix: Refactor RapidOcrModel to support all PP-OCR languages via version/backbone resolution

[nikos-livathinos]

Reworks how RapidOcrModel maps a user's requested language and backend onto concrete PP-OCR checkpoints, so Docling can serve every language RapidOCR supports instead of just chinese/english.

Previously the model shipped hand-maintained per-language, per-backend model-path tables that were hard to extend and effectively capped support at a couple of languages. This PR replaces them with a resolver that picks the right PP-OCR version and recognizer for a given (lang, backend) pair and fetches assets on demand.

Resolution procedure:

1. Get the user's input for the language and the rapidocr backend (or the default values).
2. Use RapidOCR/python/rapidocr/utils/model_resolver.py to check if the language is supported in PPOCRv6:
   a. First pass the input language through the COMMON_LANG_ALIASES.get(input_lang, input_lang)
   b. Check if the output of a exists in PP_OCRV6_LANGS
   • If yes download the PPOCR-v6 checkpoints. Return.
3. Else if the backend is torch:
   a. If the input language is in: arabic, cyrillic, devanagari, ka, korean, latin, ta, te
   download the PPOCR-v4 checkpoints. Return.
   b. Else raise exception that the asked language is not supported by that backbone
4. Else:
   a. If the language input is any of: arabic, ch, cyrillic, devanagari, el, en, eslav, korean, latin, ta, te, th
   download the PPOCR-v5 checkpoints. Return.
   b. Else raise exception that the asked language is not supported by that backbone

RapidOCR model artifacts download:

1. If the user has NOT provided artifacts_path the downloading is fully delegated to RapidOCR lib:
   • This allows to use the artifacts bundled in the python package.
2. Otherwise RapidOcrModel implementation downloads the artifacts:
   • The actual URL is still provided by RapidOCR.InferSession
   • Docling's code decides the saving path and filename.

In case the user has provided custom paths for each individual model (det/rec/cls), they are used instead.

Checklist:

• Documentation has been updated, if necessary.
• Examples have been added, if necessary.
• Tests have been added, if necessary.

[github-actions]

✅ DCO Check Passed

Thanks @nikos-livathinos, all your commits are properly signed off. 🎉

[mergify]

Merge Protections

🟢 Merge protection satisfied — ready to merge.

Show 1 satisfied protection

🟢 Enforce conventional commit

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?:

[codecov]

Codecov Report

❌ Patch coverage is 98.24561% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
docling/models/stages/ocr/rapid_ocr_model.py	98.21%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!