p:choose/p:try and sequence outputs

Following the discussion on the comments list, I think the proposal is
to allow p:choose and p:try to have subpiplines where there are
outputs that are sometimes a sequence and sometimes a singleton, with
the semantic that a sequence is allowed if any subpipeline specifies a
sequence.

And to add a primary result document to p:error which always produces
an empty sequence.

I have reservations about making these changes without going back to
last call. I'm prepared to be persuaded, but...

                                        Be seeing you,
                                          norm

-- 
Norman Walsh <ndw@nwalsh.com> | When we reduce our own liberties to
http://nwalsh.com/            | stop terrorism, the terrorists have
                              | already won.-- Reverius