ENH: Add support for dict show_inherited_class_members

[larsoner]

Over in MNE-Python we subclass some standard types like list and dict (I know, not a recommended practice) and for these few cases we don't want to show inherited class members. For all other classes, we do.

This PR adds support for numpydoc_show_inherited_class_members as a dict allowing mapping on an individual-class-name basis. If people agree this is a reasonable way to go, I'll add a test then it should be good to go.

I was hoping just adding :no-inherited-members: to a new autosummary template class_no_inherited_members.rst and using it for the few classes where I need it would be enough, but I don't see a way to get/see this switch inside NumpyDoc.

[larsoner]

For an example in the wild, this is our current docstring with a bunch of dict stuff we don't want:

https://mne.tools/dev/generated/mne.Forward.html

And this is it with no dict stuff (from mne-tools/mne-python#10911 using this branch):

[larsoner]

And we actually have a use case where we want to document some subclass methods (from our own mixin classes) but not others (from subclassing dict)... but I'm not sure a good API for this, or how to get autosummary to play nicely with it.

[rossbar]

I'll admit I have to look up how autosummary/autodoc work together every time I think about them so I may be thinking about this completely wrong, but is there a way to generate the stubs via autoclass (which already has a no-inherited-members option) and have autosummary pick up those generated stubs instead of generating them itself?

I'm just trying to think of ways to do this with the existing machinery, but I'm no sphinx expert so maybe it isn't possible (or is extra hacky). The approach taken here in conjunction with mne-tools/mne-python#10911 seems like a good way to go about this, so if IYO this is the best way then I think this is reasonable.

[larsoner]

I tried messing with autoclass options and Jinja templates for some time but couldn't make it work that way

[larsoner]

Okay since there are some approvals I'll go ahead and merge, thanks for looking @jarrodmillman @rossbar !