[RFC][WIP] Auto map vars

[stuart-cls]

Continuing from #654 (I'll rebase when ready)

Users get very confused (and I find it annoying) when Hyperspectra plots vs (clusters, map_x) or something instead of just (map_x, map_y) when they are present.

I added a XYDomainModel which reliably plots XY data in Hyperspectra if both these vars are present (otherwise reverts to existing behaviour).

If this works, I'll also apply to the other mapping-aware widgets (Bin, Stack, Polar, etc)

[stuart-cls]

@markotoplak I saw your comment in #765 (comment) and wondered if my approach here with a custom XYDomainModel would also suffer from the settings bug you reference?

[markotoplak]

Yes, it could IF that model is connected to something that is changing a context setting AND that context setting is not limited in the same way as the model.

[stuart-cls]

Hmm for part one, definitely:

class ImagePlot(BasicImagePlot):
    attr_x = ContextSetting(None, exclude_attributes=True)
    attr_y = ContextSetting(None, exclude_attributes=True)

For part two, indeed they seem to have different definitions: exclude_attributes is not exactly the same as METAS | CLASSES with valid_types=ContinousVariable

[stuart-cls]

I see some tests are failing now :D I'll dig more

[stuart-cls]

Indeed this triggers biolab/orange3#6721 in TestOWHyper.test_few_samples
It also hits a new bug since I wrote it due to ae1df1e since PyListModel.Seperator doesn't have a .name attribute.
And it's a bit weird that the map_x/map_y variables end up duplicated.

So I'm starting to think it's not worth the complexity, especially now that the attr_x/y combo boxes aren't hidden.