pg_upgrade bug in Windows on PG 9.2

I just got a bug report from EnterpriseDB saying pg_upgrade generates a
file share violation on PG 9.2.  I was initially confused because I know
we fixed this in a May commit.  Well, it turns out that this commit
re-added the same failure:

    commit 4741e9afb93f0d769655b2d18c2b73b86f281010
    Author: Alvaro Herrera <alvherre@alvh.no-ip.org>
    Date:   Thu Jun 28 23:27:00 2012 -0400

        Make the pg_upgrade log files contain actual commands

        Now the log file not only contains the output from commands executed by
        system(), but also what command it was in the first place.  This
        arrangement makes debugging a lot simpler.

The commit was certainly a good idea, but exec_prog's log_file
specification was designed as for use in an error string, meaning that
the creation on an error string at one call sight would cause pg_upgrade
to erroneously create a log file named "file1 or file2" on Windows.  The
bigger problem is that the log file was opened at the start of the
function (with no error return check), and kept open until the end,
causing the file to be opened while the command string was run --- this
was causing the share violation.

I have applied the attached patch to head and 9.2 to fix all these
issues.

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

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