Rexec rexecd dpopen

[JianyuWang0623]

Summary

This PR contains two fixes for the rexecd/rexec utilities under netutils/.

1. netutils/rexecd: use dpopen() instead of popen() to get raw fd
   popen() now returns a fopencookie()-backed FILE*. On such a stream
   fileno() returns the cookie pointer instead of a real file descriptor,
   which is typically negative. rexecd's relay loop feeds that value into
   poll(), where a negative fd is silently ignored, so the loop never sees
   the child's POLLIN/POLLHUP/EOF and blocks forever waiting for the command
   to finish. (Related upstream change: system/popen: Avoid copying FILE #3511.)
2. netutils/rexec: initialize ret to avoid maybe-uninitialized

Impact

• Users: rexecd no longer hangs after the remote command finishes; the
  command-completion (EOF/POLLHUP) path works again. No API or CLI change.
• Build: removes a potential -Werror=maybe-uninitialized build break in
  rexec.c.
• Dependencies: rexecd now depends on dpopen()/dpclose()
  (CONFIG_SYSTEM_POPEN), the descriptor-based counterparts of
  popen()/pclose().
• Hardware / Security / Compatibility: no change. Documentation: none needed.

Testing

Built with the standard nuttx-apps toolchain; warning-as-error build of
netutils/rexec and netutils/rexecd passes after the ret initialization fix.

Runtime verified by running rexecd on the target and issuing a remote command
from rexec:

• Before: rexec <host> <cmd> printed the command output but then hung; the
  session never returned to the prompt because poll() ignored the negative
  fileno() and the relay loop never observed command completion.
• After: the same rexec <host> <cmd> prints the output, receives EOF/POLLHUP
  when the child exits, and the session terminates cleanly.