Add helper function Tensor::names() (#24914)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/24914
There are two helpers, Tensor::names(), and Tensor::opt_names().
- Tensor::names() always returns a DimnameList; if the tensor doesn't have
names, it returns a DimnameList of all `None` names.
- Tensor::opt_names() returns an optional<DimnameList>: it returns
names if the tensor has names allocated, otherwise, nullopt.
Tensor::opt_names() is more of an implementation detail. It is
recommended that devs use Tensor::has_names() and Tensor::names()
because those result in a cleaner API.
This PR also cleans up callsites of Tensor::opt_names() to use
Tensor::names() where applicable.
Finally, this PR also adds impl::get_names(TensorImpl*), which is the
analogous function for TensorImpl*. (Tensor::opt_names() <->
impl::get_opt_names(TensorImpl*)).
Test Plan: - run existing tests. [namedtensor ci]
Differential Revision: D16919767
Pulled By: zou3519
fbshipit-source-id: ef30c9427a3d8e978d2e6d01c7f74f5174ccd52c
