Partially resolve #257: video_test

[smcv]

PR Description

This brings py-sdl2 a bit closer to fully resolving #257, by reducing the number of places where it assumes SDL_GetError != '' implies that there was an error. #267, #269, etc. are targeted versions of this addressing places where a newer SDL2 release does break that assumption, whereas this PR is a preventative version addressing places where a newer SDL2 release might break that assumption but has not done so yet.

This PR only covers video_test, treating all other tests as out-of-scope. Many other test modules will eventually need a change similar to this one.

• video_test: Don't check error unless a function failed

  Helps: Incorrect use of SDL_GetError() to check whether calls failed #257

• video_test: Use _check_error_msg() a bit more often

• video_test: Mitigate unsuitability of SDL_GetError() for detecting failure

  SDL_GetError() is like errno: it's documented not to be suitable for
  detecting failure, only for getting more details if failure was already
  detected (its result is unspecified on success, because a successful
  API call might have been implemented by doing something that failed,
  detecting that, and falling back to doing something different).
  However, some functions in SDL2 return void, so we have no other way
  to tell whether they have failed (they do return a result in SDL3).

  To make it less likely that upgrading SDL2 will make these tests regress,
  clear the error indicator immediately before calling the function under
  test. It is still not guaranteed to be empty on success, but at least
  this way we make sure it doesn't already contain an error message from
  a previous function call.

  Helps: Incorrect use of SDL_GetError() to check whether calls failed #257

Merge Checklist

• the PR has been reviewed and all comments are resolved
• all CI checks pass
• (if applicable): the PR description includes the phrase closes #<issue-number> to automatically close an issue
• (if applicable): bug fixes, new features, or API changes are documented in news.rst

[a-hurst]

Thanks again, I'll try combing through the other test modules and fixing the other spots I missed!

I also like the convention of documenting the checks that aren't 100% safe for the functions that return void, I'll try to keep that up elsewhere too.