Adds figure test capability

[jeffreypaul15]

The figure tests can be performed using verify_cache_image() method in test_plotting.
3 command line arguments can be passed

• reset_image_cache : rests the image_cache directory with all the new figures
• add_image_cache : option I added to make sure we don't mistakenly add a new figure unless this is specified to be True
• ignore_image_cache : ignores the image cache, doesn't perform comparing of the images.

The comparison happens through pyvista.compare_images() and the a threshold is set for the allowed error
(IMAGE_REGRESSION_ERROR >= 500 and warning at >=200).
High variance tests allow for a higher threshold with IMAGE_REGRESSION_ERROR >= 1000, but we haven't used this as all of our tests fall in the normal range (probably because we're not changing lighting conditions). I've left this in as maybe some of our tests might fall in this range.

Usage -

plotter.show(before_close_callback=verify_cache_image)
# plotter being PyvistaPlotter not SunpyPlotter

[jeffreypaul15]

That's odd, it seems to be working fine on the azure pipelines

[jeffreypaul15]

There seems to be some issue with the tests working on pytest and not working with tox.
The image is giving me an error of 12145 but after taking a look at the images, they seem to be the exact same though.

[dstansby]

So I think we just need to either reduce the tolerance of the plot_quadrangle test, or upload a new reference image. I'm not sure how we can get at the image generated on Azure pipelines though...

[jeffreypaul15]

@nabobalis @dstansby Would it be better to test each mesh separately or test them plotted a map.
Also would it be okay to have a test for each example to make sure that they're producing the right output?

[dstansby]

In general it would be good to keep the number of figure tests to a minimum, as that then reduces the number of test images we have to store (and potentially change in the future). We can test multiple features by adding them to the same test, and just making sure they're in view of the camera (ie. visible in the saved image).

At the moment, I think we should have two tests:

• One for everything except pfsspy field lines
• One for pfsspy field lines (including the map used to generate them)

[dstansby]

Also, since I've just merged #41 you'll need to rebase this to make sure the tests don't error due to any new warnings.

[jeffreypaul15]

In general it would be good to keep the number of figure tests to a minimum, as that then reduces the number of test images we have to store (and potentially change in the future). We can test multiple features by adding them to the same test, and just making sure they're in view of the camera (ie. visible in the saved image).

At the moment, I think we should have two tests:

* One for everything except `pfsspy` field lines

* One for `pfsspy` field lines (including the map used to generate them)

Alright, I shall update them asap.

[nabobalis]

SQUASH MERGE THIS