Implement a Scalar type?

[Jakob-Unfried]

Strategy

class Scalar(SymmetricTensor):
    # a tensor with no legs, i.e.~a scalar
    codomain = []
    domain = []
    ...

Reasoning

• Should be able to deal with backend-specific scalars (e.g. numpy scalar, torch scalar).
  In particular, should be able to have the scalars on GPU, so we dont need to sync hardware just to do e.g.
  S = S / tensors.norm(S)
• Packaging scalars in a tensor class allows a consistent and backend-agnostic interface (e.g. dont need to import torch to do conj)