Tests cleanup

[mileslucas]

This cleans up a lot of the unit tests, fixes regressions in the travis script as well as fixing read the docs auto-deployment. I managed to extremely bungle my git branches so this has one commit with a lot of changes from a manual reset.

Changelist:

tests

• removed tests that weren't tests (profiling scripts that didn't do much) or just outputted plots instead of running actual assertions
• removed random data and plots
• renamed existing tests to fit pytest discovery standard (filename test_*.py classname Test* and test methods test_*)
• changed the tests to be more in line with pytest standards especially the use of fixtures
• created a pytest fixture for downloading phoenix spectra and verified it works locally
• moved what seems like semi-permanent test data that isn't too large into a tests/data folder

.travis.yml:

• fixed build matrix and now test on 3.6 and 3.7
• simplified steps to just install pipenv dev packages and the the inplace cython build
• caching appropriate folders (like tests/data to reduce integration test time
• verified tests run on travis successfully (they do see here)

Note I ran into the same issues with timing out on the phoenix downloads. I opened a support ticket with travis ci to see if they had any ideas what was happening. To remedy this, in the meantime, I have marked the tests requiring this files to skip until it is resolved. I can confirm that the tests run locally as written. (#108)

read the docs

• I have added a badge to README.md that shows the doc build status
• I have made some small changes to the sphinx conf.py to make sure the RTD build would succeed
• I have tested that this build auto-deploys to RTD correctly There are some configuration settings that ought to be set for RTD to make sure it looks for the right branches and publishes them. I can list them it if you want or you can add me as a contributor on RTD.

Some larger picture things-

I have decided to make the minimum required python version 3.6. I have two main reasons for this-

1. I kept experiencing some really awkward regression failures on python 3.3-3.5, which required solutions that would greatly increase the complexity of simple operations like file saving
2. Python 3.6 includes format strings, which would help clean up a lot of the code, if we can enforce python >= 3.6 we can take advantage of this across the code.

I don't think this kind of change should be that big of an issue. I've never ran into an issue with requiring 3.6 in my other projects and if people are installing from pip it will all be handled for us automatically anyways. I've written changes into the README.md and docs that show this change as well, for consistency.

[iancze]

This looks fantastic, thanks! Happy to add you as a contributor to the RTD, though I had trouble finding mileslucas there. Do you have a different username?

[mileslucas]

I had an old handle on a non-professional email, I just made a new account under mileslucas so give it a try again.