Allow pg_dumpall to use dumpmem.c functions, simplify exit code

Bruce Momjian wrote:
> > I was wondering if it wouldn't make more sense to have pg_dumpall supply
> > its own version of exit_horribly to avoid separate pg_malloc and
> > pg_strdup ... but then those routines are so tiny that it hardly makes a
> > difference.
> > 
> > Another thing I wondered when seeing the original commit is the fact
> > that the old code passed the AH to exit_horribly in some places, whereas
> > the new one simply uses NULL.
...
> 
> I am thinking we should just get rid of the whole AH passing.
> 
> I have always felt the pg_dump code is overly complex, and this is
> confirming my suspicion.

I have developed the attached patch which accomplishes this.  I was also
able to move the write_msg function into dumputils.c (which is linked to
pg_dumpall), which allows pg_dumpall to use the new dumpmem.c functions,
and I removed its private ones.

FYI, I found write_msg() was a useless valist trampoline so I removed
the trampoline code and renamed _write_msg() to write_msg().  I also
modified the MSVC code.

-- 
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + It's impossible for everything to be true. +