Batch TrOCR line inference with configurable batch size

[sushan5140]

Closes #6

Summary

• Batch cropped line images before calling the TrOCR processor/model
• Add a configurable --batch-size CLI option (default: 4)
• Preserve line ordering and the existing NLP correction step
• Keep batch_size=1 as the previous single-line inference behavior
• Keep custom transcribe_fn behavior unchanged for tests/integrations

Tests

• Adds a focused test verifying 5 crops with batch_size=2 produce batches [2, 2, 1]
• Verifies decoded line order is preserved
• Validates invalid batch sizes are rejected

Scope

Only the cropped-line inference stage is batched; segmentation and export behavior remain unchanged.

[estherg12]

Hi!
Thank you very much for collaborating with the project. I hope to see you again around here (:

[sushan5140]

Hi! Thank you very much for collaborating with the project. I hope to see you again around here (:
Thank you so much! Really enjoyed contributing to your project. I'd love to stay connected and exchange ideas sometime. Looking forward to collaborating again!