For the Chinese characters, have you already excluded character encoding problems?

For ksh, there should be translations: https://translatewiki.net/w/i.php?title=Special:MessageGroupStats&language=ksh&group=out-wikimedia-mobile-wikipedia-android-strings Maybe the issue is that some contain comments in <!-- -->.

I cleaned up a lot of unsupported HTML tags in the ksh translation but I didn't find any HTML comments. It's still exporting just <?xml version="1.0"?>. For zh-hans, zh-hant, and nan, the export encoding is UTF-8 which is what's expected but perhaps I'm missing your point. I also looked for malformed HTML in these translations but didn't notice any.

The empty ksh file is still being generated but values-x-invalidLanguageCode has not been generated lately.

Warning: Entity 'nbsp' not defined in Entity, line: 2

Do html entities need special handling in this file format, or this just an issue in the way I use DOMDocument?

Change 331681 had a related patch set uploaded (by Nikerabbit):
Handle ampersands in translations

https://gerrit.wikimedia.org/r/331681

AFAIK HTML entities must be escaped unless they appear within <![CDATA[ and ]]

Change 331681 merged by jenkins-bot:
AndroidXmlFFS: Handle ampersands in translations

https://gerrit.wikimedia.org/r/331681