PocoMC missing numpy transform for model output

[SBouchard01]

When passing the result of an acm.observables.CombinedObservable instance model, the output of get_prediction is a numpy array, while the output of FCN.get_prediction is a torch Tensor. This causes observation-prediction to raise a TypeError for unsupported operation in the PocoMC sampler.
A quick fix is to detect the type of the prediction and turn it back to a numpy array if it is a Tensor