Fix FFDetrDetector crash when final batch has one page

[Teekyy]

Fixes #30

Summary

Fixes a crash in FFDetrDetector.extract_widgets() when the final batch contains a single page (e.g. len(pages) % batch_size == 1 and len(pages) > 1).

Root cause

model.predict() returns:

• List[sv.Detections] for batch size > 1
• sv.Detections for batch size == 1

When the final batch size is 1, results.extend(predictions) iterates the sv.Detections object and adds non-detection items (e.g. tuples) into results, which later breaks .with_nms().

Fix

• Use append for single sv.Detections and extend for lists (ensures results is always List[sv.Detections]).

Testing

• Confirmed the crash on main with a 4-page PDF using batch_size=3 (batches of 3 + 1).
• Confirmed fixed with this change; also tested batch_size=1.