The XeLaTeX backend for the new OCG PDF renderer does not fallback if the font selected for a given language does not contain a given codepoint. For many possible fonts for a given language, the latin code pages are not included. This makes page numbers, dates, citation numbers, and even bullets in lists render as tofu (blank square boxes).
The latex renderer should keep track of which code pages are present in a font, and add explicit font-switch commands to the output when needed. (Including redefining the command used for bullets in lists to ensure it is rendered in a latin font.)
Ideally we could automatically generate coverage tables from a font. But at *least* we should treat the Latin codepoints as a special case, and fall back to the default latin font when latin codepoints are used in a font without latin codepage coverage.
(This has been an issue for Russian, Indic languages, the google noto fonts, etc, and accounts for most of the present difficultly in choosing an appropriate font for a particular wiki language.)