While the automated tests didn't and couldn't catch the MuPDF issue on some systems (like Android) issue, it's unclear to me why they didn't catch this wider issue with DjVu.
These two tests open a DjVu file and they don't fail, even though it did in the actual program.
I nuked the borked v2019.1 release files and I'll replace them with v2019.1.1 tonight.
I'll hopefully also find some time to write a test that'll properly fail on v2019.1 to prevent future regressions like this.