[Feature] Handle the Device Index

Description

In its current state, the aidge_learning code completely ignores the CUDA device index.

We propose to address this significant lack in this MR.