This fixes a very tricky bug as well: the clones store the shift
internally, so when we transform a source layer, we should transform
the clone's offset as well. So we need to represent our transformation
with a matrix to do this. Now there is a matrix and a test for it in
KisTransformWorker.

WARNING: there is a bug(?) in the worker. Its transformation differs 1-3
pixels from the theoretical values. Looks like there are some rounding
problem. Anyway, we can use it even in such a rought way.

The bug will be closed when the visitor will be added to KisImage
CCBUG:280502