Integrate prediction head into model's forward pass for loss_fn

[clementpoiret]

Ensure the loss_fn in hackathon/objectdetection.py correctly utilizes the newly implemented prediction head by calling eqx.combine(params, static) and then jax.vmap(model, ...) to get the detection logits.

[bernardev254]

I will ensure the loss_fn in hackathon/objectdetection.py correctly utilizes the new prediction head by integrating eqx.combine(params, static) and applying jax.vmap(model, ...) to obtain detection logits, optimizing the object detection process.