Skip methods that have incomplete types

[liamappelbe]

Fixes dart-lang/native#247

Also change how method return types are parsed to fix dart-lang/native#246, since I couldn't write the test otherwise. This also fixes dart-lang/native#260, fixes dart-lang/native#281, fixes dart-lang/native#267, and fixes dart-lang/native#312

[liamappelbe]

Seems to be making some method return types too generic. I'll have to investigate a bit more.

[liamappelbe]

Looks like the tests are failing to find Foundation/NSObject.h, which is weird. I don't think I've changed anything relevant to that, and I can't repro this locally. So I'm re-running CI at head to see if something has changed there: https://github.com/dart-lang/ffigen/actions/runs/2562399073

[dcharkes]

Foundation/NSObject.h

That was the same issue as I had earlier when I tried to re-order finding libclang.dylib on MacOS. Maybe something on the MacOS machines in the Github workflow changed?

Though, I could repro it locally (when trying to use libclang from XCode instead of one of the others).

[liamappelbe]

Yep, the mac bot is failing at head: https://github.com/dart-lang/ffigen/actions/runs/2562399073

Filed dart-lang/native#249

[dcharkes]

Changes are LGTM.

I'm missing some test cases for the bugs that are closed. Should we add those cases to a test to make sure we don't regress them in the future?

[liamappelbe]

I'm missing some test cases for the bugs that are closed. Should we add those cases to a test to make sure we don't regress them in the future?

They're covered by bad_method_test. They're also covered by the objc example I'm writing. Once I get that working reliably, I think it would make sense to turn that into a test. It'd be good to have some integration tests that are using real mac APIs.