Test with thread sanitizer

[romangg]

Recent CI fails because of data races under load showed that there were still many tests not behaving well in regards to thread safety (for example at merge of !5 and !6) and potentially there are many more.

While one can fix them one by one whenever a pipeline fails it would be better to check with a thread sanitizer. This would also prevent new data races from creeping in. A separate build is then needed because the thread sanitizer can't be used together with the others we currently employ.

The issue is: I tried this already and the method to check-wait with a signal spy on a signal being emitted seems to be inherently racy. It's always a read after write until the read is successful. I'm unsure how to solve that. Maybe a helper class with a mutex?

[romangg]

The Plasma Window Model test seems to be especially problematic:

FAIL!  : PlasmaWindowModelTest::testTitle() Compared values are not the same
   Loc: [/home/roman/dev/kde/wrapland/src/autotests/client/test_plasma_window_model.cpp(648)]
FAIL!  : PlasmaWindowModelTest::testAppId() Compared values are not the same
   Loc: [/home/roman/dev/kde/wrapland/src/autotests/client/test_plasma_window_model.cpp(682)]

I could reproduce this locally when running the test through valgrind and having stress running for some time.

[romangg]

Removing the discussion label again and moving forward to execute. After all the goal is clear: test with thread sanitizer. If this is realized in the end by introducing mutex around every spy or something else is a technical implementation detail and if this needs discussion the issue can be moved back.

[romangg]

This should help to find the last remaining leaks that only show very sporadically on CI and can not reproduced locally. For example: https://gitlab.com/kwinft/wrapland/-/jobs/492708632

[romangg]

mentioned in commit e44986d

[romangg]

mentioned in commit c2065ba