Make create_f_CNODE also return the NODE

[SCiarella]

In this way there is no need for the user to call directly NeuralODE(*).

Then, create_f_CNODE will need to receive as input also the parameters that will be used for the solution, like t_range, algorithm, ecc..

[luisaforozco]

If the force is needed, for example to create the loss, one could get it via: my_neuralODE.model

[luisaforozco]

This is implemented now in branch wrap_neuralODE.

However, I see that there are some sort of memory leaks (many allocations) that were not there before. We need to first understand what is going on before merging that work.

[luisaforozco]

That PR was closed. The branch should also eventually be removed but I leave it in case we want to look back at something. I close this issue because this is no longer relevant, we changed completely the way of writting the forces.