If I may weigh in on this softly as someone not actively developing (but actively watching and advertising), I would point out that while in the mathematics world referring to footnotes may not be common or even desirable, in the humanities world it's not unusual. Footnotes/endnotes often have complex constellations of citations and other information (sometimes very long!) which needs reminding of hundreds of pages later, and referring to a footnote - even in a footnote, yes - is the most efficient way to do this.
That doesn't mean PTX needs to support that behavior, per se, either in theory or in terms of developer heartburn. But design decisions may make it easier or harder, or just less likely, for certain fields to adopt PTX for certain types of publications.
If I were to express a specific preference, I think combining Oscar's idea with 3. seems plausible as a middle ground. That is to say, only allow fn xrefs in footnotes, where it would only allow display as "see footnote 6 in section 5.3". This would still sharply curtail where footnote xrefs are allowed (presumably helping development) and look a little bit more like what is typical in fields that do this more often.
<fn>I've never seen anything like "footnote 4.6.2", it's always something closer to "see footnote 5" - also depending on whether footnotes are numbered continuously or by chapter. See any Supreme Court decision for references by page number and footnote of the style "Contra, ante, at 21, n. 5.", once again often within the footnotes. Even more extreme, see [1] for an example of a style for referencing a footnote in *another* work.</fn>
<aside>Endnotes seem to be getting much more popular nowadays, particularly with "popular" books that still want proper connections to the literature, but also in scholarly texts. I'm not presumptuous enough to suggest PTX implementing endnotes, which would probably cause more than just heartburn, but just FYI that I could imagine an actual author requesting this someday.</aside>