POSIX: don't leak FDs when launching child Processes.

* On POSIX, make sure we don't leak FDs when launching child Processes.
* Add a facility to LaunchProcess() to remap a given FD into a child process.

This change is needed for 2 reasons:
1)We want to use a socketpair() for IPC, the child process needs a known FD # for it's side of the socket.
2)The OS X Sandbox doesn't close FDs.

[jeremy, 2008-12-17 00:41:46]

Also re-enable process_utils_unittest.cc on OS X.

Mike: Did we disable the test for some reason or did we just forget to enable
it?

[agl, 2008-12-17 01:50:15]

LGTM

http://codereview.chromium.org/14497/diff/10/14
File base/process_util_linux.cc (right):

http://codereview.chromium.org/14497/diff/10/14#newcode69
Line 69: close(src_fd);
No need to close here, you've already marked it close-on-exec. Also, it
precludes something like [(5, 1), (5, 2)] (both stdout and stderr going
somewhere)

http://codereview.chromium.org/14497/diff/10/15
File base/process_util_mac.mm (right):

http://codereview.chromium.org/14497/diff/10/15#newcode60
Line 60: if (posix_spawn_file_actions_adddup2(&file_actions, src_fd, dest_fd) !=
0)
I don't know with Mac stuff, I assume that it's correct :)

[pink (ping after 24hrs), 2008-12-17 19:09:05]

LGTM

http://codereview.chromium.org/14497/diff/10/16
File base/process_util_posix.cc (right):

http://codereview.chromium.org/14497/diff/10/16#newcode41
Line 41: // rlim_t is a uint64 - clip to maxint.
why?