implement YOLOv10-style loss function with bbox, objectness, and clas…

[bernardev254]

Summary

This PR adds a composite loss function for YOLOv10-style object detection, combining box regression, objectness, and classification losses.

Changes

• Implements loss_fn() in hackathon/objectdetection.py.
• Computes bounding box loss (L1), objectness loss (BCE), and class loss (BCE).
• Weights losses with configurable λ parameters.
• Returns both total loss and individual components for monitoring.

Motivation

This loss function is essential for training the YOLO-style model and is consistent with the design goals of balancing localization, confidence, and classification accuracy.

closes #5