RFC: Running both modeltester implementations in parallel

In #239, I added support for the C++ Qt modeltester, so that the Python implementation is only used when the C++ one from Qt is not available. Nowadays, the C++ implementation is available for all backends, so the Python one is... rather useless. Of course I only realized after already porting the newest changes 😅

With #420, we now have some additional checks which only landed in Qt 6 (but already work and are useful for e.g. Qt 5.15 too), and with #421, we get checks which aren't even in a Qt release yet (they will be in 6.4.0 in around 5-6 months).

We've been favoring the C++ checker because it can check some stuff we can't (better access to internals), but with that, this now isn't as true anymore (especially since people might want to check older Qt releases, thus also getting an older C++ checker). We also get much nicer output due to pytest's assertion handling.

I'm wondering: Should we deprecate force_py and always run both checkers in parallel? The obvious drawback is somewhat verbose/redundant output (both Python and C++ checker, though in different sections).

Review issues #239, #420, and #421 first, then trace how the force_py option selects the Python and C++ modeltesters. The issue does not name files or tests; done would require a maintainer decision on parallel execution and deprecation before implementation can be scoped.