subpicture: render text region in a separate picture region
...or into no region for Text To Speech renderers.
On top of !4493 (merged), only the last commits matter
We no longer need the ugly hack that marks the output region as marked and we don't mix the input and output regions.
Edited by Steve Lhomme