Join on multiple keys

[astrofrog]

I realized today that we can solve the linking case described in #958 by extending join_on_key to support linking on pairs, triplets, and n-tuples of components rather than just single components. The current PR implements a very rough (and very slow) version of that, just to test the concept. If things work well, then I can work on a way to make the calculation of the masks much faster. Here's an example in use:

We have a position-position-velocity cube and a second moment map (on the same celestial pixel scale).

from glue.core import DataCollection, Data
from glue.app.qt.application import GlueApplication
from glue.core.data_factories import load_data
from glue.viewers.image.qt.viewer_widget import ImageWidget
from glue.viewers.histogram.qt.viewer_widget import HistogramWidget

data_linewidth = load_data('linewidth.fits')
data_cube = load_data('ppv_cube.fits')

dc = DataCollection([data_linewidth, data_cube])

# We can now link on pairs of components
data_linewidth.join_on_key(data_cube, ('Pixel x', 'Pixel y'), ('Pixel Axis 2', 'Pixel Axis 1'))

ga = GlueApplication(dc)

image = ga.new_data_viewer(ImageWidget)
image.add_data(data_cube)

histogram = ga.new_data_viewer(HistogramWidget)
histogram.add_data(data_linewidth)

ga.exec_()

Selecting in the linewidth histogram will now highlight pixels in the image.

Note that the fact the pixel axes have different naming conventions needs to be fixed (see #959)

@ChrisBeaumont - do you have any thoughts on this approach so far?

[astrofrog]

Note that at the moment this is not going to be backward-compatible in terms of loading save files.