It looks like one of the users managed to catch such a case,
when the node has been deleted while the updates were still
holding a pointer to it. I don't know how it was reproduced,
and, technically, our system should not allow such cases.

Therefore, this patch just converts the node pointer into a
weak pointer and adds a sanity check assert into KisAsyncMerger.

I hope we'll catch this assert one day and get better report
about what triggered it.

CCBUG:405879