align train_step with new object dict structure and loss input format

[bernardev254]

Summary

This PR updates the train_step() function in hackathon/objectdetection.py to properly align with the new object detection loss function and batch structure.

Changes

• Extracts and converts image data to NCHW format.
• Unpacks detection targets from the updated batch["objects"] structure.
• Passes YOLO-style detection targets to loss_fn, including bbox_target, object_mask, and class_target.
• Computes gradients and updates model parameters using Optax.

Motivation

This brings train_step() in sync with the newly implemented loss and evaluation logic, enabling full-cycle model training for object detection using YOLOv10-style predictions.

closes #8