Simplify tests: use stdlib, adjust to organization conventions

[diego-plan9]

Summary

Revise the tests so they only use stdlib features, not requiring additional dependencies in order to be executed. Tweak folders and add docstrings and linting adjusting to the rest of the organization conventions. Update the tox configuration accordingly, using unittest directly for running tests, along with pycodestyle and pylint for linting, in the same spirit as terra, aer and the rest of the organization repositories.

[mtreinish]

I'm fine with using stdlib unittest for the tests here we have now, being 2 very quick tests just to sanity check. I only used stestr here to be consistent with what we were doing in terra. If we don't care about that, then we can wait until parallelizing the tests actually makes a difference.

As for pylint I really feel that pylint is far too pedantic and actually impedes productivity instead of improving it by enforcing consistent style (besides not working on py37 properly which is a personal pain). This patch alone is a good example of this things like the docstring rules and other changes made so that pylint will pass don't actually add anything to the code or make it more readable. While the only code in the metapackage is tests I don't think we should force this just because other repos are using it. Especially if we're not going to make other parts of validation consistent with other repos. I also have an open issue about this on terra too: Qiskit/qiskit#1179 to discuss moving away from pylint over there too.

[diego-plan9]

As mention in the description, it's about consistency: I can understand your views on pylint, but I really don't think pointing to an open issue (which was actually informally discussed in a meeting and my recollection is that we were leaning towards not moving away from it) and personal preferences justify deviating from the current practices we have in the project.

[mtreinish]

Except as I said this doesn't actually make things consistent, the test runner used with this patch is now different from terra. We either care about consistency or we don't. In this specific case I don't think it really buys us much tbh. It's also not like we're consistent across all python projects in the qiskit organization repos. For example, aqua doesn't run any style enforcement, pylint or otherwise, in their ci and other repos have differing rule sets. Also we don't pin pylint versions across repos that do use it. Differing versions of pylint behave very differently.

That all being said we're also only talking about test code (and not a lot of it) here which is fundamentally different from everything else you're comparing against. Enforcing things like every module, class, and function having a docstring does not add anything to test code. (Personally I don't think that kind of enforcement adds much to lib code either, but that is a different argument)

The only reason I mentioned the open issue is to point out this not a settled discussion.

[delapuente]

As @mtreinish says, let's delay the use of stestr for when parallel tests make a difference which reduces the complexity.

On the other hand, the use of pylint is consistent with the qiskit-terra codebase. This look great to me. Merging.