Simplify testing framework

[climbfuji]

Description

For capgen, we build and run the tests with CMake, and if it passes then we run another Python-based test on the output. That seems redundant and unnecessarily complicated to me.

Welcoming feedback on whether this is a good idea or not.

Does the Python framework test aspects of the API that the CMake tests don't test? Or are the CMake ctest-based tests comprehensive enough to do away with the Python-based tests?

What we should avoid is testing frameworks that are redundant. For example, every time a variable changes in the output list, one has to fix the Fortran test code and the Python test code.

Solution

Consolidate the testing frameworks as appropriate so that there is no redundancy, unless absolutely needed.

Alternatives (optional)

Do nothing

Related to (optional)

#737

[climbfuji]

Update. I now recall that we talked about this a while ago. We figured that SIMA is using the Python API (ccpp_datafile.py), that the other host models should also be using the Python API when transitioning to capgen, and that we should rather consider dropping other methods to query the capgen datatable.