Consider caching charset subset relation

I have code here that computes the FcCharSetIsSubset relations for all fonts in the current configuration, saves that data to a file, and loads it again:

https://gitlab.gnome.org/GNOME/pango/-/merge_requests/229

That makes coverage trimming much cheaper, since most of the charset comparisons can be avoided. The code probably belongs into fontconfig, but I don't have the energy to rewrite it and figure out how to integrate it into fc-cache.

It is slighty tricky, since it is global information, not per-directory.