Add EasyOCR wrapper with rotation correction

[M-O-J-U]

What does this PR do?

Adds ocr.py: wraps EasyOCR, tries all four right-angle rotations and keeps the best-scoring result, supports configurable languages per document type.

Which module(s) does it touch?

docling_pk/ocr.py. Depends on docling_pk/preprocessor.py.

How did you test it?

Tested against a real photo shot sideways: unusable output before rotation correction, correct output after. Also tested en+ur versus English-only on certificate images, since certificates are almost entirely English text and the combined model was introducing Arabic-script noise into English words.

Edge cases handled:

No detections returns empty text and 0.0 confidence rather than raising. Language list is configurable per call, defaults to en+ur.

Checklist:

• Tests added or updated
• Docstrings on all new public functions
• No unused imports
• README updated if public API changed (no public API change, internal module)
• Lead Engineer has reviewed and approved before submitting to director (no separate Lead on this project currently, Khubaib and Khadija are aware and approved proceeding solo)

Note: this PR was pushed earlier but the GitHub PR itself was never opened. This is the actual fix for the current CI failure (docling_pk.ocr missing from main). Recommend reviewing and merging this before PR #9 (Copilot's auto-fix).