PR: Define Qt/binding versions at top level, fix warnings if versions not found, and fix test dir on CIs

[CAM-Gerlach]

Followup to merging PR #289 to master following a very weird and unexpected check failure on Win 10 + PyQt 5.9 + conda + Python 3.6; see CI logs.

Seems like there's some weird behavior on this very specific CI run, where the tests pass, but when we run the command to save the path for the test coverage, which imports QtPy a second time in a new instance of the same interpreter, QT_VERSION is not defined, which must mean that all the if blocks either didn't trigger or had an ImportError and QtPy fell back to using PyQt5. I could understand if PyQt 5.9 didn't define those top-level constants (since the source repo is not public, I tried to check beforehand but couldn't), but the fact that it worked without error the first time for the tests leads me to believe that it must either have been something with setting the env variables that triggered the different behavior on the second, or there's something else I'm not accounting for going on.

EDIT: It was the latter, the second Python command was run without -I which resulted in shadowing due to the hack I used to avoid problems due to the non-src directory layout. See below for full details.

While this PR fixes the proximate error and ensures that the top-level constants are always at least defined, I want to see if I can reproduce this locally in the same env, as it may indicate an underlying issue. Done, see below.

So, this PR implements three levels of fixes to the related issues discovered here:

• Checks that QT_VERSION is defined first before acting on it (like all the others)
• Initializes the top-level version constants to None at the top of the file, which avoids them being undefined if no Qt API is found and also reduces duplication and verbosity
• Fixes the shadowing issue in the tests that surfaced this behavior to begin with

[CAM-Gerlach]

Okay, I figured out what's going on. Since QtPy doesn't use the recommended src directory layout, in the CI tests I had to cd into another directory, in this case qtpy, to avoid testing the local copy rather than the actual installed package (which not only would not detect any packaging problems, but also plays havoc with the coverage). However, this was a rather short-sighted fix, as this results in QtPy's sip module shadowing the sip package (and could result in other such tricky issues). This doesn't cause problems for the tests since I pass the -I flag to python in the tests command, which prevents Python from adding the working dir to the path. However, when the standalone command is run to import qtpy and get its path for the coverage, importing PyQt5 fails due to this and thus version is left undefined. I'll fix that as well here, to avoid any future knock-on effects from this, by just cding into a temporary directory instead.

[CAM-Gerlach]

@dalthviz @ccordoba12 if either of you could look this over and merge this if there's nothing serious going on, so that tests are passing on master again, that would be great, thanks.

[ccordoba12]

Looks good to me, thanks @CAM-Gerlach!

[CAM-Gerlach]

Thanks @ccordoba12 !