Add Dimname, DimnameList to codegen to support named tensor build
I'm working on enabling named tensors by default (see
https://github.com/pytorch/pytorch/pull/25181). As a part of this
process, I am removing the build flag that was guarding the feature.
This isn't a BC-breaking change: named tensors still won't work with XLA
but unnamed tensors should still work.
Removing that build flag exposes some named tensor only data structures
(Dimname, DimnameList) that can be arguments to new operators and now
become present in the xla codegen. This PR fixes the namespacing for
those.
