Potentially we fixed a crash when the mouse is being released,
because the interaction strategy is not guaranteed to exist
at that time.

There is still a crash happening, namely the simutaneous
access to some KoShape. It seems that there is some scheduled
updating running in the GUI thread outside a stroke
catching the race condition. One possible place is
KoShapeManager::Private::updateTree(). It can be either called
directly or by updateTreeCompressor.timeout(). The latter one
is presumably unsafe.

Task: T10901