feat: OCR evaluator

[cau-git]

Adds an OCR evaluator which computes CER.

Picked out and updated from branch integrate-ocr-benchmarks, see #46

⚠️ This code is currently not fitting the docling-eval API and must be re-written.

TODO

• Rewrite OCREvaluator to fit the BaseEvaluator interface (especially arguments and return types of __call__)
• Match the design of the other evaluator classes. No output_dir, and no writing to the file system anywhere.
• Remove custom code for reading from JSONL, and custom visualization
• Update and test with the generic visualize facility (see here).
• Prove that it works as intended with completed test unit. It must be able to run at least with PixParse OCR dataset and docling predictions (see feat: PixParse OCR dataset builder #61)

[samiuc]

@cau-git I've completed all the to-do list items you mentioned in the description above. Just a quick note, we now calculate both CER and Character Accuracy. I've also added the tests for each hyperscaler for OCR and the code now follows the same design pattern as the other modules.

[cau-git]

@samiuc thanks for checking off the TODOs!
Below are a few remaining remarks from my end.

[PeterStaar-IBM]

LGTM!