Substantial reworking of the font rendering code to measure
the size of the text before rendering, thereby creating correctly-
sized output images (rather than ones with guesses on height and
with the full target width).
This uses some potentially dubious behaviour described here
https://stackoverflow.com/questions/
31788577/create-pango-layout-before-cairo-surface
and could cause any number of problems.
The advantages are that a) it should be a little more efficient and
b) the images created for the subtitle analysis code are useful.