Change `Tensor` check in `autojac.jac`

[PierreQuinton]

Currently we check with isinstance, in torch they do this:
https://github.com/pytorch/pytorch/blob/449b1768410104d3ed79d3bcfe4ba1d65c7f22c0/torch/autograd/__init__.py#L448

Which is slightly different, we may want to use the is_tensor_like function from torch to remain compatible.

[ValerianRey]

I agree. Let's do that + change all other checks of isinstance(something, Tensor), if any.

[PierreQuinton]

It is defined as

def is_tensor_like(inp):
    return type(inp) is torch.Tensor or hasattr(inp, "__torch_function__") 

This means that the object is either a Tensor, or has implemented some amount of torch operations (similar to torch_dispatch we used in SLTs)

[ValerianRey]

Can you make a PR for this? Otherwise I'll do it

[PierreQuinton]

Can and did