Refactor QA setup

[aleksihakli]

• Utilize tox with tox-travis for more UX friendly test orchestration
• Utilize pytest with pytest-django and pytest-cov for more readable test output
• Utilize setuptools_scm for inferring package version
• Improve test matrix definitions for database and Python version combinations
• Skip false-positive test for cProfiling output

[nasirhjafri]

@aleksihakli Just two things, we dropped Django 1.11 support in the last release and we are still supporting python 3.5

[aleksihakli]

@nasirhjafri @auvipy @jezdez the django-silk QA improvements are ready for upstreaming. There is one false-positive case in the tests in the cProfile section that results from a small difference in timings with Python 3.6 interpreter, but otherwise all tests pass and the test bench is considerably improved.

The test also fails on PyPy3 in addition to Python 3.6 and should be refactored to account for differing cProfile outputs or disabled altogether if it is not necessary:

[nasirhjafri]

LGMT 👍 Thanks @aleksihakli

[jezdez]

Some minor updates needed..

[jezdez]

Hmm, the test should be written differently then.

[aleksihakli]

I very much agree: however, the test is not simple to fix as the result is a more complex object and ideally the format should be validated, not the contents itself.

Hence, the test would ideally use some sort of a regex matcher for a matching list that is inside the list of profiling results.

This test and the max body size test are also the ones that fail on PyPy3 at the moment.

I would opt a skip in this PR and then a fix in a later PR that updates the test to a more appropriate format, but I am not interested in refactoring the tests at this moment due to time constraints.

[aleksihakli]

@jezdez I believe the version is ready for merging now. Thanks for the accurate review insights!

[jezdez]

Actually, some test is failing now, not sure why?

[aleksihakli]

Seems like it's the same test. For whatever reason it isn't skipping the test with the @unittest.skipIf decorator. I think I'll refactor the test after all to validate the format of data instead of it's exact output so that the test supports PyPy as well.

[aleksihakli]

Test refactored to test output format instead of specific tool output, seems to be working. Upstreaming the changeset.