Use resolved values for subscript and superscript defaults where needed

[Hoolean]

Happy new year!

---

A few of the superscript and subscript properties are unusual in that they depend on the resolved value of other attributes: e.g., superscript size should fallback to the explicit subscript size if provided, not just share its default.

This is hopefully an easy +1 for Noto Serif Oriya, which is now happily building at 100% locally : )

Worth noting

• The PR adds test coverage for an Oriya-adjacent case, although we could add more to be certain there's no regression
• The second commit does some early rounding to match ufo2ft's precision loss -- where further transforms occur, we need to round as early as ufo2ft does to match the exact final value in all cases -- but this wasn't based on a real-world use-case, and so we could drop it if we prefer simplicity unless compatibility demands it

[Hoolean]

Oof, just realised this duplicates #1823, sorry @cmyr 🙊... there's some parallel universe stuff here like using the closure return to avoid get(...), and the optional rounding commit, so feel free to Frankenstein hack and splice this one if anything ends up useful