REGRESS_OPTS versus MSVC build scripts

I believe (but won't be able to prove for another few hours) that the
reason the MSVC buildfarm members are failing on contrib/unaccent,
as seen here:
http://www.pgbuildfarm.org/cgi-bin/show_log.pl?nm=mastodon&dt=2009-08-18%2013:00:00
is that they are not running the test in a UTF8-encoded database.
And the reason for *that* is that the MSVC build infrastructure pays no
attention to REGRESS_OPTS.  Instead it's got hardwired values for all
the pg_regress command line options, and of course it needs to hardwire
several different sets of values for different tests :-(.

I think it's time to fix that properly by making the perl scripts
dredge REGRESS_OPTS out of the makefiles.  My perl-fu is not up to
the task, however.  Anybody want to do it?

This seems like a stop-ship issue for 8.5alpha1, even though it's only
a contrib regression test that's certain to fail; so I think it's a
bit urgent.
        regards, tom lane