4919: beam 1/20 and shorter notes by 1/4 in 2/2 and 3/2 time

4919: beam 1/20 and shorter notes by 1/4 in 2/2 and 3/2 time

Else 10 or more notes end up on a beam by default,
which is hard to read and unusual.

[dak, 2016-07-04 17:55:57]

Since this is supposed to be an improvement in typesetting, it would seem rather
appropriate to include a regtest demonstrating intended use.  That makes it easy
to consider the improvements achieved for the same input.

[simon.albrecht, 2016-07-04 18:45:05]

Add regtest coverage.

[dak, 2016-07-04 20:21:41]

Ok, so this example makes clear the effect in 2/2 and 3/2 meters.  Thing is, I
am not really convinced with the result in relation to the meter.  Do we have
any word from Gould and/or Ross about this situation?

[dak, 2016-07-22 06:20:32]

https://codereview.appspot.com/303980043/diff/20001/scm/time-signature-settin...
File scm/time-signature-settings.scm (right):

https://codereview.appspot.com/303980043/diff/20001/scm/time-signature-settin...
scm/time-signature-settings.scm:73: ((beamExceptions . ((end . ((1/20 . (5 5 5
5))))))))
Well, the other exceptions tend to draw the line differently (see 6/4, 9/4,
12/4, and the somewhat different 3/4 and 4/4 resume beat-beaming at triplets
rather than smaller).  One problem with quintuplets are that they are
indistinguishable from decemtuplets in the exception rules.

Take for example

\score {
  \fixed c'
  {
    \time 2/2
    \tuplet 10/8 2 { c16 d e g f d g f e d
		     c g e d f e g f d e }
    c1
  }

  \layout {}
  \midi { \tempo 2 = 60 }
}

under old and proposed rules.  So I think that when changing the
autobeaming rules, we should try to keep LilyPond's defaults somewhat
consistent and predictable and not arbitrarily draw the tuplet lines
differently for every meter.

I particularly don't see why quintuplets at 3/2 should be treated differently
than at 6/4 as compared to triplets.