Updating test suite to yt-4.x and concurrent testing

[jwise77]

This PR updates all testing infrastructure to be compatible with yt-4.x, including

• fields as tuples
• removing calls to the halo mass function
• referring to yt_astro_analysis for HaloCatalog calls
• removing yt.mods calls
• replacing na with np (!)

I created a new gold standard (v15) because the old gold-standard cannot be compared to the new runs without some changes to the testing environment (e.g. different packages and dependencies). I manually confirmed on my local machine that the latest PR (#132) passed the push suite before making the new gold standard.

Another big change was using process pools in the multiprocessing python module to run the test simulations concurrently. This saves about 10 minutes in the push suite simulations. I couldn't get the testing to run in parallel without putting more effort into it. This could be addressed in a future PR. I increased the CircleCI resource from medium to large since we can run tests in parallel now, and the large nodes are still free to us.

Note: There are a lot of duplicate changes in common code in various files, so the review should be easier than what the diff size says.

[brittonsmith]

A heroic effort, John! Thanks for doing this. This looks good to me.

[brittonsmith]

Now that the yt infrastructure we rely on is going away, we may want to think about moving to something like what we've setup for Enzo-E. That system can be parallelized automatically using circleci.

[pgrete]

Thanks for the fixing this. Looks good to me. I only have a couple of non-blocking clarification questions.

[gregbryan]

Looks good to -- just a few duplicated imports. Thanks for doing all of this!

[gregbryan]

Now that the yt infrastructure we rely on is going away, we may want to think about moving to something like what we've setup for Enzo-E. That system can be parallelized automatically using circleci.

Britton -- Can you be more specific about what exactly is going away?

[jwise77]

Now that the yt infrastructure we rely on is going away, we may want to think about moving to something like what we've setup for Enzo-E. That system can be parallelized automatically using circleci.

That's good to know about the answer testing infrastructure going away. I've made issue #192 to remind us about it.

[jwise77]

Looks good to -- just a few duplicated imports. Thanks for doing all of this!

You're welcome! I've committed your suggestions. Those duplicates were left over from my find & replace!

[brittonsmith]

Now that the yt infrastructure we rely on is going away, we may want to think about moving to something like what we've setup for Enzo-E. That system can be parallelized automatically using circleci.

Britton -- Can you be more specific about what exactly is going away?

The yt project is transitioning its answer testing from nose to pytest. The AnswerTesting class we import in test_runner.py from yt will be quite different from how it is now. I'm not sure what compatibility the new pytest framework will have with our current setup. I will investigate.

[gregbryan]

This looks ready to go -- thanks @jwise77 for fixing testing; now we can address many outstanding pull requests. I'm merging.