Fix unit-test fixture teardown

[eskaur]

We noticed that unit-tests were failing when run with SDK 2.11. It turns out that we were not tearing down objects in the right order, and SDK 2.11 is less tolerance to such things.

This commit does two things to fix the unit-tests:

• Make sure that all fixtures that yield frames end up releasing those frames on teardown.
• Fix a test that created a new Application inside its code while also indirectly depending on the "application" fixture.

[nedrebo]

We noticed that unit-tests were failing when run with SDK 2.11. It turns out that we were not tearing down objects in the right order, and SDK 2.11 is less tolerance to such things.

Maybe mention this in the SKD changelog @apartridge? Just a reminder about the lifetime requirements for Zivid::Applicaton. Probably some customer code will hit the same and then it is nice if it is already addressed in the Changelog.

[eskaur]

We noticed that unit-tests were failing when run with SDK 2.11. It turns out that we were not tearing down objects in the right order, and SDK 2.11 is less tolerance to such things.

Maybe mention this in the SKD changelog @apartridge? Just a reminder about the lifetime requirements for Zivid::Applicaton. Probably some customer code will hit the same and then it is nice if it is already addressed in the Changelog.

Honestly the docstring for Application is outrageously short

https://www.zivid.com/hubfs/softwarefiles/releases/2.6.0+fa0036c5-1/doc/cpp/classZivid_1_1Application.html#details

[apartridge]

We want to first understand why there is a regression in 2.11 compared to 2.10. We will investigate this.

[eskaur]

Actually the docstring in the header is good. But I can't find the same info on the doxygen page.