0-dim named tensors crash on _force_order

[gtanzer]

Scalar named tensors (without named dimensions), like those returned from loss functions, crash when _force_order is called internally. A minimal example is:

a = NamedTensor(torch.tensor(0.0), ())
b = NamedTensor(torch.tensor(1.0), ())
a+b

The relevant part of the trace is:

...
/usr/local/lib/python3.6/dist-packages/namedtensor/core.py in _force_order(self, names)
    195                 trans.append(d)
    196         return self.__class__(
--> 197             self.transpose(*trans)._tensor.contiguous().view(*view), ex
    198         )
    199 

/usr/local/lib/python3.6/dist-packages/namedtensor/core.py in transpose(self, *dims)
    132         )
    133         indices = [self._schema.get(d) for d in to_dims]
--> 134         tensor = self._tensor.permute(*indices)
    135         return self.__class__(tensor, to_dims)
    136 

TypeError: permute() missing 1 required positional arguments: "dims"