Making the abyss geometry handling work with babl_process_rows is proving a bit
tricky - the choice that skips the skipping at the start of a batch entering a
tile seem to let all expected code paths work and no reported non-working code
paths yet. This will also be a slight performance loss - but it might be that
we seldom do writes that fall in the abyss anyways.

Not writing into the abyss will permit better control over sliced rendering
with multi-threading - and should be added back later.