Return type of .Norm()

[pcchen]

Now

• Tensor.Norm() returns a rank-0 Tensor.
• UniTensor.Norm() returns a rank-0 UniTensor. Does it make more sense to return a number?

numpy.linalg.norm retuns float, unless you use axis argument to compute vector norms. But I don't think we need to support axis argument.

[manuschneider]

It would make sense to me to return double instead of rank-0 tensors. Was this maybe intended to allow for different data types and not return the same precision for every type? Anyways, I would suggest to change it. But it would break existing code.