[stdlib] Rename `VectorProtocol.Scalar` to `VectorSpaceScalar`. (#25374)
A step towards machine learning optimizer revamp.
The next step is to make `Tensor.VectorSpaceScalar = Float`.
This makes `Tensor` scalar multiplication always work with `Float`
instead of `Scalar`.
Renaming `VectorProtocol.Scalar` is necessary. Otherwise, `Scalar`
is ambiguous within `Tensor`'s type context: it may refer to either:
- The witness type of `VectorProtocol.Scalar`, or
- The `Scalar` generic parameter.