Update tests to actually treat 'flask' as optional

[sirosen]

1. Add a tox factor, flask, for installing flask via the flask extra
2. Update the tox matrix to run with and without -flask in addition to -minimum_flask
3. Make test collection ignore the flask/ test dir if flask can't be imported (also, rename the matching tests to tests/flask/)
4. Minor changes in those tests so that the conftest.py won't fail during test collection

---

There's a long-standing inconsistency in that the tests have always required flask even though the library declares it to be an optional dependency.
We may soon have an actual usage of this library (an AP CLI client) which doesn't need flask, so resolving this one way or the other will help ensure that installs don't have broken imports if flask is missing.

[kurtmckee]

I think I was expecting some tests to be marked with a skipif or skipunlessimport (or whatever it's called), or to have a custom pytest marker associated with the Flask tests.

Is that what the collect_ignore in conftest.py is doing? If so, then I've learned a very neat new thing!

[kurtmckee]

What does assert flask do?

[sirosen]

Oh, I did that on the fly for flake8's F401. Let me replace with a noqa comment.

[sirosen]

Yeah, I tried out the collect_ignore instead of a skip marker or similar because I wanted to avoid even trying to load those modules (and dealing with their imports). It's new to me, and we can obviously always switch to a skip marker if we find it doesn't work well.

One interesting thing is that even with that collect option set, the conftest.py in the relevant subdirectory gets loaded. To me, that was a big surprise! So I wouldn't call this a "new and recommended pattern" yet -- just a different way of doing things which seemed to fit this use case nicely.