Parse double flats in RomanText

[malcolmsailor]

Fixes #1390.

For the vast majority of cases (where there is just a single flat symbol) the regex on line 83 is overkill so we only run that if the preceding conditions don't match.

[coveralls]

Coverage increased (+0.0003%) to 93.066% when pulling 3847da9 on malcolmsailor:double-flats into ac6d6d2 on cuthbertLab:master.

[mscuthbert]

Also a test for a different double-flat key than b, say 'Ebb' since that follows a completely different code path potentially than 'bbb'

[malcolmsailor]

I think both cases follow the same path since the checks for B/b/Bb/bb only apply when the key has two or fewer characters. But I added such a test.

I wasn't sure what the right balance between comprehensiveness and readability was here since these are actually doctests (albeit ones that people who aren't actually developing music21 are probably unlikely to consult).

[mscuthbert]

A few small changes to make the common case of Key of 'b' or 'B' work more quickly.

[malcolmsailor]

OK, done. Rather than checking for 'B' explicitly I just pass if the last character is anything other than 'b'; then we don't check all those other redundant conditions if e.g., the key is "F#" or "E", etc.

[malcolmsailor]

It just occurred to me that this function will throw an exception if textString is the empty string. Should we handle that case or is the exception ok?

[mscuthbert]

Looks great! Thanks @malcolmsailor and for the last patch @jacobtylerwalls We can make empty string throw an exception later, but if so we should do it explicitly and not just with an IndexError which would be strange.