Ensure arbitrary adjoined elements render correctly in matplotlib

[philippjfr]

Currently you can adjoin any element type to another and at least in bokeh they will render correctly in general. However in matplotlib the dimensions of the adjoined plots will often be wrong due to issues with aspects. Additionally many matplotlib plots do not correctly support invert_axes, which leads to the data being plotted incorrectly in those cases. These two issues should be addressed.

[jbednar]

Will that require changes in matplotlib itself? If so may be worth opening an issue there now so that they are aware of it, even if we don't plan to work on it at our end right away.

[philippjfr]

No, I think this will just require going through all matplotlib plotting classes and making sure that it behaves correctly when adjoined.

[jbednar]

How will that help fix invert_axes?

[philippjfr]

The plotting classes should invert the data appropriately so invert_axes is respected. The AdjointLayoutPlot will then tell the subplot that it should invert the axes if it is adjoined to the right. That's how it works in bokeh and for the few plot types that support it in matplotlib.