Not there yet on regression-test DST independence

The regression tests now pass okay in PST season.  However, I took what
in hindsight is an obvious precaution: I set my system clock forward to
next summer and tried them.  In PDT season we still have a problem (see
attached diff).  Will leave it to you to select the most appropriate fix.

While I was at it, I tried setting the clock forward to winter and
summer of 2025, and got the same regression-test behavior as for now.
So at least we don't have any near-term "oops, this date is now in the
past" dependencies, like the one that bit us last June.
        regards, tom lane


*** ./expected/horology.out    Fri Oct 19 21:02:21 2001
--- ./results/horology.out    Sun Jul  7 11:21:19 2002
***************
*** 555,561 ****   + interval '1 month 04:01' as timestamp without time zone) AS time) AS "07:31:00";  07:31:00
----------
!  07:31:00 (1 row)  SELECT interval '04:30' - time with time zone '01:02-05' AS "20:32:00-05";
--- 555,561 ----   + interval '1 month 04:01' as timestamp without time zone) AS time) AS "07:31:00";  07:31:00
----------
!  08:31:00 (1 row)  SELECT interval '04:30' - time with time zone '01:02-05' AS "20:32:00-05";

======================================================================