Issue 4205: Improve part combiner's rest analysis

Issue 4205: Improve part combiner's rest analysis

Rests must begin and end simultaneously to be merged into the shared
voice.

Rests, skips, and multi-measure rests are kept apart even if they
begin and end simultaneously.

This does not produce ideal output in every case, but it avoids
producing musical nonsense.

[Dan Eble, 2014-11-26 04:30:57]

Simplify and add regtest

[Keith, 2015-05-13 04:55:23]

I noticed the same bug with rests and skips.  I'm using the patch below with
success on my own music, but I haven't tried it on the regtest suite yet.

https://codereview.appspot.com/174610043/diff/20001/scm/part-combiner.scm
File scm/part-combiner.scm (right):

https://codereview.appspot.com/174610043/diff/20001/scm/part-combiner.scm#new...
scm/part-combiner.scm:51: (filter f? (events vs)))
I suggest also sorting rests before skips, so that we compare the rest to the
rest if parts contain a mix

 (append (filter (lambda(x) (ly:in-event-class? x 'rest-event))
                 (events vs))
         (filter (lambda(x) (ly:in-event-class? x 'skip-event))
                 (events vs))))

https://codereview.appspot.com/174610043/diff/20001/scm/part-combiner.scm#new...
scm/part-combiner.scm:457: (= 1 (length rests2))
(< 0 (length rests1))
(< 0 (length rests2))

[Dan Eble, 2015-05-17 02:55:47]

On 2015/05/13 04:55:23, Keith wrote:
> I noticed the same bug with rests and skips.  I'm using the patch below with
> success on my own music, but I haven't tried it on the regtest suite yet.

Thanks.  You've given me a push, but I think this calls for some more
significant changes to handle same-part multi-measure rests and skips as well. 
I'll work on it.