fixed up romanText repeats

[malcolmsailor]

This PR addresses much of #858, though not (I think) the case of mid-measure repeats.

It turns out that romanText repeats were already largely implemented (by whoever wrote the romanText/translate.py, presumably), but there were a couple bugs that were preventing it from working as expected and there was no test.

I wrote some test cases and debugged the existing code. The bugs are explained in comments in the code. As far as I can tell, the repeats now work smoothly.

I also have a few (non-urgent, I think) questions in the comments. I'll repeat them here:

1. repeat.Expander does not work on stream.Stream, only on stream.Part. Is this expected behavior?
2. in the case of 1st and 2nd endings, spanner.RepeatBracket.getNumberList does not return the measure numbers I expect. What is the correct approach here?

[malcolmsailor]

pylint objects to a bunch of stuff in translate.py but I don't believe any of it is introduced by this PR

[coveralls]

Coverage increased (+0.01%) to 93.034% when pulling f8f72a4 on malcolmsailor:roman_repeats into 1a0ec50 on cuthbertLab:master.

[mscuthbert]

yes, the problems were introduced in this PR when additional modules were imported at the top of the file. romanText.translate cannot import converter, because romanText is itself a converter, so we'd get a circular import.

[malcolmsailor]

Oops, sorry about that, right you are. It slipped my mind that I had imported converter. I should have been more careful and linted the code without my changes.

[mscuthbert]

I'm going to merge this, since three of the issues are small (style/grammar) changes in test code, and the one substantive one (the TODO) is described carefully and can be a separate issue. I'm concerned that getNumberList() is returning false information, but that seems like either a misunderstanding of all of us, or a separate bug.