Review mathml-entities and char-def and rethink where the data goes

[alerque]

Hoisted from #2167 because I need to think about it and there doesn't seem to be a reason to hold up that show:

By the way, it obviously has some overlaps with our char-def.lua file used for linebreaking (notably), and if I understand it correctly, "stolen" from ConTeXt by Simons years ago. I haven't checked how those could be reconciled...

Originally posted by @Omikhleia in #2167 (comment)

The data being used from upstream Unicode XML Entities (specifically in our usage mathml-entities) and the stuff in char-def could both probably use re-homing. That might be in cldr-lua, potentially a ucd-lua package, and or others.

Just as another point of review, we may not need Lua packages for this. There may be Rust packages out there (e.g. ICU family stuff, codex, and others) that has this data too.