This patch adds a shrinking pass that tries natural text
transformations - unicode decomposition (NFD/NFKD) and case mapping -
on individual characters in string choices. Failures involving e.g.
""À" != "À".lower()" will now reliably shrink to ""A"" rather than
sometimes getting stuck on the high-codepoint accented form (issue
#4725).
The canonical version of these notes (with links) is on readthedocs.