implement preprocessing logic for object detection samples

[bernardev254]

Summary

This PR implements the preprocessing() function for the object detection pipeline in hackathon/objectdetection.py.
closes #1

Changes

• Extracts image, bounding boxes, and labels from VOC-style TFDS samples.
• Resizes images using prepare_image.
• Converts bounding boxes from absolute coordinates to normalized format [ymin, xmin, ymax, xmax].
• Flattens the "objects" dictionary into separate "bboxes" and "labels" fields.
• Returns a structure compatible with downstream augmentation steps.

Motivation

This change enables the object detection pipeline to correctly prepare samples for augmentation and model input. It aligns the preprocessing output format with the expectations in the augmentations() function.