Split kerning by script, not by direction (third attempt)

[madig]

This is a clean-up PR for #667.

This implements:

• Making classifyGlyphs also deal with tuples instead of just lists and sets.
• Making the KernFeatureWriter emit scripts independently of the globally declared languagesystems. This is arguably a separate concern. I removed some tests that no longer make sense.
  • The code now looks at the declared languagesystems and the font's codepoints that have just one script associated with them, as a heuristic for what the font is supposed to support. We don't depend on the designer to add all desired script declarations.
• Having KerningPair deal with just pair information, and deal with scripts and bidi metadata outside. This simplifies the code a lot by collecting concerns in one place.
• A method of not triggering erroneous subtable breaks in otlLib (swallowing kerning) by:
  • Using a simpler kern splitter that passes kerning validation and produces no overlapping kerning groups
  • Sorting the kerning pairs (glyph-to-glyph first, then glyph-to-class, class-to-glyph and class-to-class)
• Some more tests to test bidi and script splitting
• Dropping kerning classes from being written to the feature file
• Various code simplifications of the KernFeatureWriter.

Note that for reasons I have not investigated, various Noto Serif Tamil sources have some errant kerning, but they have the same problem with the current mainline kern writer.

Closes https://github.com/googlefonts/ufo2ft/issues/658

[madig]

• [x] also consider declared languagesystem, make union with the automatically found ones.

[anthrotype]

while reading old issues, I stumbled on a long comment of mine that details the way the ufo2ft kernFeatureWriter worked:

https://github.com/unified-font-object/ufo-spec/issues/16#issuecomment-488751554

It would be nice to have a similar write up for the newly porposed split-by-scripts algorithm..