Add comprehensive validation testing

[clyne]

With the inclusion of a Python API, it is now possible to add automated, comprehensive validation testing to verify that renderers are behaving correctly with different data inputs. A validation system would exercise as many combinations of renderers and data sets as possible. Validation might be performed by comparing rendered results with "ground truth" using an image comparison such as SSIM. Ground truth images might be generated by a one-time, manual inspection of images with a known working version of the software.

[clyne]

We agreed to do this in two parts:

1. A script that will input a config file (e.g. yaml) that specifies data sets, variables (2d and 3d), and a list of renderers to invoke, and will then output rendered images for each combination of data/variable/render
2. An automated validation component that would use some kind of image comparison metric (e..g. SSIM) to compare "ground truth" with images generated by the script.

Part (1) is a 3.7 milestone. Part (2) will be deferred until later.

[sgpearse]

Closed with #3191.

[clyne]

Re-opening this but removing the milestone: (1) above is done, but (2) is not.

[StasJ]

This should also resolve #2986