Category guessing leads to bizarre results

Why aren't my marks attaching to my bases? I have a base glyph "brm_RR" and that works fine, and another base glyph "brm_R" and marks don't attach. Turns out it's because:

```IPython
In [4]: from glyphsLib.glyphdata import get_glyph

In [5]: get_glyph("brm_RR")
Out[5]: Glyph(name='brm_RR', production_name='brm_RR', unicode=None, category=None, subCategory=None, script=None, description=None)

In [6]: get_glyph("brm_R")
Out[6]: Glyph(name='brm_R', production_name='brm_R', unicode=None, category='Letter', subCategory='Ligature', script=None, description=None)
```

`brm_R` goes into ligatures not bases in GDEF, so marks aren't attached. But why is `brm_R` considered a ligature? Because...

https://github.com/googlefonts/glyphsLib/blob/10c8b1d9d7a6e71871a91f2ffbb4ce5b7ccbd582/Lib/glyphsLib/glyphdata.py#L245-L247

`brm_` is skipped, R is looked up, R is a letter, and so `_translate_category` is called, and inside `_translate_category`:

https://github.com/googlefonts/glyphsLib/blob/10c8b1d9d7a6e71871a91f2ffbb4ce5b7ccbd582/Lib/glyphsLib/glyphdata.py#L295-L298

This is **way too much magic**.

I literally told Glyphs that it was Category=Letter, but no, glyphsLib thinks it knows better.

Read Lib/glyphsLib/glyphdata.py around the linked lines and reproduce the get_glyph("brm_RR") and get_glyph("brm_R") calls. Trace how the name is categorized and review existing tests; done means the reported category behavior is resolved and covered by a regression test.

Written by the indexing model from the issue text.