Remove duplicates in documenting trait members

[rkdarst]

• When run with the :members:, then traitlets traits are duplicated,
  because they are added to the autodoc list from both from
  :members: and this autodetection. (I think)
• Discussed in at least
  Jupyter(Hub) conceptual intro jupyterhub#2726. Currently
  causing JupyterHub docs to fail, because sphinx gives an error if
  there are duplicate autodoced traits and it is run with -W.
• This seems to be started in a new version of sphinx, but we aren't
  completly sure. JH has been using the -W option since 2017.
• I'm unsure if this is the right solution, but it works and gets me
  past these errors.

[rkdarst]

This needs proper testing. I edited live source files to test, then copied to here. Beware!

[betatim]

If this fixes the problem in the JupyterHub docs build I would vote for merging this. My hunch is this is one of the packages that mostly only JupyterHub (or the JupyterHub ecosystem) uses. So if we break it more than it is already broken we will notice.

Let's see if Carol has time and thoughts.

[willingc]

This change seems reasonable to me.

I made some general comments in #3021.

• The docs are building on RTD as of 2 weeks ago after Georgiana's change
• If CI is failing due to the warnings, I would remove -W in CI.
• Makes sense to me to convert to rst if @minrk agrees
• I would also try to switch from autodoc to sphinx-autoapi.
• Consider Sphinx 3 in the near future too ;-)

[willingc]

LGTM. Though I did not build locally for this review.

[rkdarst]

I"ve been testing with a manual pip install https://... from JH, and it works. Last night, I thought that the sorting/ordering might need adjusting (if it's not sorted in the consumer), but... having the trait_members first actually sort of makes sense, so I guess it's not a big deal.

My concern about trivial syntax errors is resolved now, so if a minimal test is enough, I'd say this is OK to go.

[betatim]

Merged. We are riding the elevator UPwards in the rabbit hole :)