complete evaluation by computing mAP on validation batches

[bernardev254]

Summary

This PR completes the evaluate() function by implementing mAP calculation on the validation set using model predictions.

Changes

• Combines model with eqx.combine and runs inference using jax.vmap.
• Applies postprocessing to convert raw outputs into bounding boxes, labels, and scores.
• Computes mean Average Precision (mAP) using IoU-based matching.
• Displays mAP per epoch during validation.

Motivation

Adding mAP evaluation enables quantitative tracking of model performance during training and validation, aligning with the YOLOv10-inspired object detection framework.

closes #7