Support unit suffixes

[rsheeter]

We should be getting fea unit suffixes. I would like to use these in fontc. I particularly want normalized as it is the most natural way for me to emit kerning value records. googlefonts/fontc#345 WIP emits things like and will probably block on this capability:

feature kern {
  pos dze-cy Te-cy (wght=-1n:-69 wght=0n:-61 wght=1n:-49);
  ...etc...

""u" for user units, which as I said would be the default if omitted. Design units would be "d" and normalized units would be "n"

harfbuzz/boring-expansion-spec#94 (comment).

[cmyr]

Some additional thoughts on this:

• suffixes are only valid when parsing metrics, not other numerical values
• are suffixes valid in non-variable fonts? we don't know if we have variation info during parsing, so this will be a question for the validation pass
• this also implies/requires that metrics can be non-integer ('float'/real) values. Do we accept user coordinates as non-integers, or does the presence of a non-integer metric require a suffix?

as a first pass I would say:

• just allow reals in all cases
• truncate in resolution where required
• error in validation if we see design or normalized units and don't have a variation model

@rsheeter one thought just occurred to me, which I don't have enough context to answer: will this actually simplify fontc? As far as I understand, the simplification here is that when fontc has normalized units it won't be required to convert these to user coordinates. But it will still need to provide the logic to normalize existing user coordinates, and will now have to support normalizing design coordinates as well. Is this a win?

• fontc will still need to be able to handle normalizing user coordinates
• but fontc will also now need to be able to normalize design coordinates as well

[rsheeter]

fontc benefits in two ways:

1. Generating fea with normalized units is more natural because that's what we have so we don't have to convert to user
2. We dont' have to convert the user units that we produced from normalized units back to normalized units again via callback so a bit less work is actually done iiuc

[cmyr]

added in #203