warning: touch manuals doesn't force a completely healthy rebuild

[Graham Percival]

Guys,

I broke git master by accidentally pushing
fb964a541b56576fd929739e36835ac9bd0de2b9 , which I was still
reviewing.  I'm testing a fix for those problems right now.

However, while investigating the problem, I've discovered that if a
doc build fails, a second doc build (after modifying the file *and*
touching the main texi file) doesn't regenerate the @lilypond if
you've only changed the [options].  I remember that we changed the
comparison to only look at self_ly() instead of full_ly() in
lilypond-book for some good reason... but that has the unintended
effect of sometimes not rebuilding a failed @lilypond.

I don't see it being worth playing games with this, though.  We
shouldn't be changing [options] very often anyway -- I mean, 90%
should be [quote,verbatim,relative=2], and 9% should be
[quote,verbatim].  Unfortunately there's a bunch of old cruft that
_doesn't_ follow those, so we have some short-term danger.

People testing doc patches which change [options] before pushing them
to master should probably do a build from scratch just to be safe.

Cheers,
- Graham