Here was use get_ink_extents() rather than get_pixel_size() to
find out how big the rendered subtitle will be, then use the
x/y values of this extents rectangle to offset the rendering
within the image. This allows the removal of some hacks to make
accents visible.
I'm not sure if this goes far enough - it seems impossible to find all
the font IDs used in a FFmpeg file without scanning it. I don't know
if (for example) SSA-format font IDs make it through to DoM when decoded
by FFmpeg.
