That's fair on the regression, and the ticket is the right route for it. If it used to work at that size, something changed on their side.
On the font work, one thing that bit us hard when we added accented glyphs to a pixel font. The font was generated from one shared table, but each build also carried its own hand typed copy of the alphabet and searched it using the generated character count. Adding nine glyphs moved that count everywhere, so the lookup ran past the end of the older arrays. It didn't crash. It printed a wrong glyph, most often for the space, and only on some screens.
Worth checking that everything indexing your character set reads its length from the same source the font is built from, rather than from a constant that was correct before the new languages went in.
If Windows and Linux embed the font separately, compare the built bytes. Ours both compiled fine while one was wrong.