Current regression tests

I've run the regression tests on today's source tree, and found lots of
ordering differences (not a problem, since order is not guaranteed), two
test failures because I changed an error message, and one different
result.

The different result is in the "select_distinct_on" test; the original
result had 8 rows and the new result has 40 rows. However, I'm getting
myself confused on what the correct result _should_ be, since "select
distinct on" is not documented. For a query like:

  SELECT DISTINCT ON string4 two, string4, ten FROM temp;

What is the "ON string4 two" clause saying? Anyway, the result is
different than before, so we would probably want to look at it. I'm away
'til after the weekend, but can help after that.

I also had the (seemingly) great idea to sort each regression result
before diff'ing, so that we are not sensitive to insignificant order
changes in output. The problem is that we then would not be testing
"ORDER BY" clauses, so we would need an extra qualifier on each test
saying whether it is OK to sort or whether it needs to stay raw.

Oh, I've got a UNION regression test ready to go; will apply next week.
Looks good Bruce!

                                                          - Tom

------------------------------