Timezone code, one more try

Ok, here is another attempt at the timezone code. (let's hope it's not
too big to get through the patches list this time) Changes from last
time:

* Now used on all platforms, not just win32 (per request)
* Removes the part of the timezone library we don't use...
* Removes support for the system tz library, *requires* the new builtin
one (per request)
* As a bonus of this, removes a bunch of workarounds around platform
specific issues
* Attempts to pick up the OS timezone upon startup (will work sometimes,
fail sometimes, because there really is no API to pick up the name of
the current timezone..). Yes, this is certainly somewhat hackish due to
this..
* Cleans out the timezone librarys use of K&R function headers, changing
them to ANSI
* As a bonus, "show timezone" will now work on an installation that does
not use the TZ env variable and does not have timezone specified in
postgresql.conf. Previously, it just returned "unknown".

For the time being, the general APIs to the timezone functions have not
been changed. We might want to do that eventually, but I figured getting
it to work with the current APIs is a good point to start.
The only exception to this is that pg_tzset() (replacement for tzset)
now returns a boolean value indicating if it has successfully changed
the timezone. That got rid of a lot of workaround codes around the lack
of failure indication.

The patch is very large, but the bulk of it is due to:
* The changing of the headers K&R -> ANSI
* The replacement of all "struct tm" to "struct pg_tm" (and the same for
the tz functions, but those are not as many)
* The replacement of all "include <time.h>" with #include "pg_time.h"

In doing this, I have also removed platform-specific regression test
outputs. Since they all use the lib now, it should be fine.

Attached files are:
timezone2.patch.gz - well, the patch.
pg_time.h - goes in src/include
strftime.c - goes in src/timezone

The following files should be removed:
src/timezone/difftime.c
src/timezone/asctime.c
src/test/regress/expected/abstime-solaris-1947.out
src/test/regress/expected/horology-no-DST-before-1970.out
src/test/regress/expected/horology-solaris-1947.out
src/test/regress/expected/tinterval-solaris-1947.out

I have tested the patch, and it passes regression tests + some other
manual tests on Win32, Linux (Slackware 9) and FreeBSD (5.2), all on
i386 hardware. I'm sure there will be *some* issues on others...

Comments?

//Magnus
<<timezone2.patch.gz>>