Porting #2450 (test job updates) to yt-4.0

[matthewturk]

I did my best in porting #2450 to yt-4.0. Let's see if it succeeds! There may be additional bits I missed that were not explicitly included in 2450, however.

[matthewturk]

Looks like some errors I need to deal with. Will do that and push a new set of commits with the updates.

[matthewturk]

I removed NUMPY_VERSION from appveyor.yml as it's not in master, and seems to break appveyor.

[matthewturk]

OK, so, it's now failing with 31 actual errors, which I believe are all real. At this point it seems that we're stuck with checking in new answers eventually, so let me check a few things out before proceeding.

[matthewturk]

Basically, where this is at: the test differences are real but not significant (pixel-level shifts), but I'm reluctant to check in yet another set of new tests until we figure out how to do so without bloating the repo too much. I'm hoping this will be resolved through something like git-lfs, but until this is discussed on the yt-dev mailing list I'm not sure.

[matthewturk]

So the plan now is to see the current travis test failures, validate the "answer results," and if they are valid, to commit them to https://github.com/yt-project/answer-store/ in the branch yt-4.0 and re-trigger a build here to make sure they work.

[matthewturk]

@Xarthisius I'm not sure I totally get the codecov results, but I think maybe it's because we're not testing the same number of versions? If that's indeed the case, is it OK if we merge this?

[Xarthisius]

@Xarthisius I'm not sure I totally get the codecov results, but I think maybe it's because we're not testing the same number of versions? If that's indeed the case, is it OK if we merge this?

The missing coverage is from not running answer tests on jenkins. Yes, it's ok to merge.