windows progname / regression fixes

Attached are 2 patches and one additional alternative contrib regression
test file (for cube). The patches fix a misleading message in
pg_regress.sh, and strip the .exe suffix from the result of
get_progname() as previously discussed. This lets us get much further
with contrib regression tests - there are 2 remaining issues preventing
a clean run - one odd one in the seg module and a path issue in
tsearch2. I am looking into these. (Why am I bothered with all this?
Because I want the buildfarm system to run on Windows).

One odd thing I have noticed in testing, is that dropping the regression
database takes a *very* long time, especially after the third or fourth
go, and I see messages about not being able to remove the directory,
which means rmtree() is failing somehow.

cheers

andrew