Fix tst_UpdateSpatialNode::updateResource race on destroy count
The test uses QQuick3DDataTest::grab(), whose shared utility now calls
window->update() and waits for QQuickWindow::frameSwapped before
invoking grabWindow(). This drives QSSGRenderNode update cycles earlier
and with less implicit event-loop spinning in the test body, exposing a
pre-existing race between the main thread and the render thread.
Backend node deletion is dispatched via cleanupResources(), which is
connected to QQuickWindow::afterFrameEnd with Qt::DirectConnection and
therefore runs on the render thread. afterFrameEnd fires after the render
phase of a frame, which can be concurrent with the main thread exiting
the QTRY_VERIFY(callCount == 3) loop that exits as soon as callCount is
set during the sync phase of the same frame. The result is that
destroyedCount may still be 0 when QCOMPARE evaluates it.
Replace QCOMPARE with QTRY_COMPARE for all three destroyedCount checks
so the test correctly waits for the asynchronous render-thread deletion
to complete rather than relying on incidental event-loop timing.
Pick-to: 6.11
Change-Id: I4d8fc89fb1dfd846427253d7819c0aff60a7d5ee
Reviewed-by: Christian Strømme <christian.stromme@qt.io>