[CI] Develop strategy for testing Tethys with and without optional dependencies installed

[swainn]

We currently build our test environment with all dependencies installed so that we can test all of the features in Tethys, even those with optional dependencies. I think we will still need to do this even after Tethys 5 is released so we can be assured that it still works with everything installed.

However, we currently don't test Tethys without the optional dependencies installed which leaves us vulnerable. This will be more of a risk in Tethys 5, so I think we should also run our test suite without all of the optional dependencies to be assured it works in that state.

I propose we run the tests twice, once with all dependencies, and once without optional dependencies. This will require us to annotate the tests somehow. Maybe an "optional_dependency" decorator that is a variation on "allowed to fail" that only allows fails under certain conditions? Like import errors or module not found errors? Starting this issue so we can discuss more.