wip markFeatureWriter: only allow Nonspacing Marks to be classified as GDEF marks

[m4rc1e]

The current method to classify marks in the GDEF table is too greedy. If a glyph contains an anchor whose name starts with an "_", it will be classed as a mark glyph. Recently, this implementation has forced us to rollback some fonts we recently pushed

google/fonts#2453

I spoke with Cosimo last week and he had the following idea for a less greedy implementation which this PR implements.

0. Implement the following heuristic only if the source files haven't specified a GDEF table
1. Get all the encoded Nonspacing Marks in the font
2. Use fontTools subset to find the unencoded Nonspacing Marks using a subset closure
3. Iterate through every glyph and every anchor in each glyph. If the anchor starts with an "_" and the anchor's glyph isn't a Nonspacing Mark, output a warning and skip it.

I'm not too familiar with the ufo2ft lib so my implementation may be incorrect/suboptimal. My idea was to simply add this functionality to the MarkFeatureWriter._getAnchorLists function since it's already filtering anchors.

Once we're happy that I'm adding this functionality in the correct place, I still need to do the following:

• Add tests
• Add doc strings
• tidyup variable names etc

[m4rc1e]

Also on the todo, fix the existing failing tests which appears to be a dumpster fire.

[madig]

Random thought: make ufo2ft or whatever print a warning when a Unicode value suggest a non-mark but there is an underscored anchor in the glyph.