What should happen if ufo's define inconsistent kerning groups?

[rsheeter]

#338 assumes kerning groups should be identical across ufos in a designspace. But ... what if that's a bad assumption? - prior art in python tries hard to handle this case by fiddling with the groups:

1. https://github.com/fonttools/fonttools/blob/4923a353bb75c0cb1f771c6f90edb63e35b3a7a2/Lib/fontTools/varLib/merger.py#L484-L508
2. https://github.com/fonttools/fonttools/blob/main/Lib/fontTools/misc/classifyTools.py#L5
3. mathKerning is wrong in many ways robotools/fontMath#22
4. Test groups for equality before doing fontMath kerning math robotools/fontMath#245

[rsheeter]

Per IM, having the groups be inconsistent reflects user or application error in handling sources. The reclassification process attempts to mitigate, but having correct (consistent groups) sources to begin with remains preferable.