mod_shell_stream: flush inherited stdio buffers in child before dup2 to stop console log leaking into the audio pipe

[asdf3309]

Description

Fixes pop/clip noise at the end of every shell_stream:// playback (see #3120).

The child branch forks from the parent while FreeSWITCH console logs (e.g. sofia's hash(insert/...last_dial/...) notices) are still sitting in the block-buffered stdout FILE* stream. The sequence:

1. fprintf(runtime.console = stdout) buffers console log text, not flushed yet
2. switch_fork() — the child inherits a copy of the buffered data
3. dup2(fds[1], STDOUT_FILENO) points stdout at the audio pipe
4. switch_system() forks a grandchild that runs system(dcmd) (the audio source); when it exit(0)s, stdio flushes the inherited stdout buffer into the audio pipe
5. The playback pump plays those log bytes as audio → tail pop/clip

This patch flushes stdout/stderr in the child before the dup2, so there is nothing buffered to leak into the pipe. The flush must happen before dup2, otherwise the logs get flushed into the pipe.

Type of Change

• Bug fix

Related Issues

Fixes #3120

Testing

• Tested manually (verified on 1.10.12 in a Docker container, 24 kHz L16 → 8 kHz alaw)

With a pure-silence repro (shell_stream://sh -c "dd if=/dev/zero bs=24000 count=4"), the pipe previously contained 1,145 bytes of log text after the 96,000 bytes of stream data, e.g.:

pipe read=1145 first=5d20736f6669612f last=0a1b5b6d   ("] sofia/..." ... "\n\x1b[m")

After the fix the pipe contains exactly the stream data, RTP capture shows a clean silence tail, and the audible pop is gone. Note this is a different failure mode than #3104 (blocking wait() in file_open): that one is addressed separately.

Note: master's printf("EOF") sentinel and 1.10.12's read()==0 end-of-stream both leave the leak path intact; flushing in the child fixes the leak for both.