patch for bug 443 (ENCODE-UNIVERSAL-TIME effectively ignores TIME-ZONE argument)

Hi, here is a patch to fix bug 443. With this change, Maxima + ABCL
passes its tests (in tests/rtest11.mac for the record) for
parse_timedate which calls ENCODE-UNIVERSAL-TIME.

Incidentally this patch also fixes an unreported bug in
DECODE-UNIVERSAL-TIME: the daylight saving flag is reported
incorrectly because EXT:GET-TIME-ZONE expects its argument in
milliseconds, not seconds.

I can't sign in to the bug tracker as the OpenID button doesn't work
and I don't remember my common-lisp.net username/password and I don't
see a way to get a password reminder on common-lisp.net.

A comment on the bug tracker mentions Java 8. If I remember correctly,
I tried the previous version of the code with both Java 7 and Java 8
and in both cases it fails. In any event the error seems to be pretty
clearly a mistake in ENCODE-UNIVERSAL-TIME so I would be surprised if
different Java versions yielded different results.

Hope this helps, and all the best.

Robert Dodier

PS. This patch was taken against r14981 on armedbear-svn/trunk.