markFeatureWriter: use GDEF to classify glyphs; make disjoint markClasses

[anthrotype]

This is towards fixing a bunch of related issues with defining what goes into the mark/mkmk feature

https://github.com/googlei18n/noto-fonts/issues/1223#issuecomment-398010907
notofonts/noto-source#122 (comment)
#261 (comment)

[anthrotype]

There are two changes in this PR:

1. If a table GDEF { GlyphClassDef @Bases, @Ligature, @Marks, ; } GDEF; is present in the features.fea, then the MarkFeatureWriter will use those groups to define what is a base glyph, what is a ligature and what is a mark glyph. When no GDEF GlyphClassDef definition is present, the current behavior doesn't change. This addresses ligature for "allah" notofonts/noto-fonts#261.

2. When a mark glyph contains multiple _-prefixed anchors, we only assign that glyph to one markClass; we no longer associate it with multiple markClasses one for each _-prefixed anchor that it may contain. Instead, we sort the glyph's anchor list by putting _bottom and _top anchors before all the rest, and then we only take the first anchor's name to define the glyph's markClass.
   This is, according to @schriftgestalt's coment in NotoSansMono-MM.glyphs fails to build... is this a source issue? notofonts/noto-source#122 (comment), what Glyphs.app does in these ambiguous cases when generating the mark feature. This is to fix a compilation error when the same mark glyph belongs to multiple mark classes that are used within the same mark lookup.