Prevent consolidateCompletedTuplets() from breaking complete tuplets

[jacobtylerwalls]

Closes #1780

Before
consolidateCompletedTuplets() was violating the Hippocractic Oath of tuplet doctors: first, break no well-formed tuplets.

In more detail:

While scanning this stream...

s = converter.parse('tinyNotation: 2/4 trip{c8 d8 e8} trip{e8 e8 r8}')

the region spanning 1.0 QL from {e8 - e8 - e8} was replaced with a single e4, even though it broke the tuplets it spanned.

The reason was that the algorithm scanned the first e8, determined that it didn't match the pitch of the immediately prior d8, and reset all of the search state.

After
Now, search state is maintained regardless of whether a note is reexpressible or whether it matches the prior note. This way, search state is always cleared when a tuplet is completed, ensuring no candidates for consolidation are identified inside other well-formed tuplets.

[coveralls]

coverage: 93.059%. remained the same
when pulling 30d9355 on jacobtylerwalls:jtw/tuplet-follow-ups
into b8306fc on cuthbertLab:master.

[jacobtylerwalls]

Happy to move this to another module if this is too exposed (or too confusing with TupletFixer).

The location in the duration module explains why you see stringified annotations or stringified gEBC() calls, because this module can't import note. So that would be ripe for a polish if there's a better location.