Turns out that 84ea33ef did not actually
fix the problem; it was merely good enough to fool the asserts in
FastSaxSerializer that were not yet adapted to the funky maMarkStack.

The problem is that the StartRuby()/EndRuby() call StartRun()/EndRun()
and that really messes up the mark stack.  Prevent that by first ending
the run before starting new one in StartRuby(), and then starting a new
run after ending the ruby-base one in EndRuby().

The hints handled in SwWW8AttrIter::OutAttrWithRange() are sorted in the
right way for this to work (i.e. the EndRuby() will be the last call for
any given position).

This apparently causes an additional <w:r/> empty element to be written,
not sure if that causes any issues but at least the output is
well-formed for fdo82849-1.odt.

Change-Id: I28ffd029246310ff0e405b0674aef8e724700d4d
(cherry picked from commit c2c51fa5)
Reviewed-on: https://gerrit.libreoffice.org/17230Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: David Tardon <dtardon@redhat.com>
Tested-by: David Tardon <dtardon@redhat.com>