Merge lp://qastaging/~afrantzis/mir/fix-surface-visibility-event-test into lp://qastaging/mir

tests: Fix race conditions affecting MirSurfaceVisibilityEvent tests

tests: Fix race conditions affecting MirSurfaceVisibilityEvent tests

Three problems/races were identified and fixed:

1. WindowManagerPolicy function calls are normally serialized using locks in BasicWindowManager. However, msh::CanonicalWindowManagerPolicy::select_active_surface() could be called unlocked through SurfaceReadyObserver, leading to various races and subsequent failures. This was fixed by serializing select_active_surface() with an internal (recursive) lock. There are other solutions, which involve changing the API, but I didn't want to interfere with the API without Alan's feedback. This solution is good enough for now.

2. For reasons unknown, googlemock doesn't fully clear the expectations with VerifyAndClearExpectations(), causing intermittent failures for visibility events received during surface teardown. This problem was worked around by unsetting the client surface event handler before surface destruction.

3. Waiting for a surface to become visible is not a sufficient condition for testing that the surface has reached a steady state as needed by the tests. We also need to wait for the surface to get the focus. This MP fixes the problem by waiting for both visibility and focus state when creating a surface. We use a spin_wait instead of monitoring surface events to workaround the same googlemock problem mentioned in (2).