Make extrapolation with NaN Unitful-friendly

[Ickaser]

Checklist

• Appropriate tests were added
• Any code changes were done in a way that does not break public API
• All documentation related to code changes were updated
• The new code follows the
  contributor guidelines, in particular the SciML Style Guide and
  COLPRAC.
• Any new documentation only uses public API

Additional context

Would in theory close #365 .

[SouthEndMusic]

Maybe also add a test to prevent regression.

[Ickaser]

Maybe also add a test to prevent regression.

Happy to do so; should I just test with Float64 or go all the way to adding Unitful as a test dependency to test the case I was working with?

[SouthEndMusic]

I think it's fine to add Unitful as a test dependency, but better ask @sathvikbhagavan first.

[sathvikbhagavan]

Yes, its fine to add it as a test dependency

[Ickaser]

OK, I've added a test for this specific case, and on my machine all the tests now pass. As far as I can tell, there were no similar NaN checks for other interpolation types, so this should be good to go I think.