Kill a remote process stuck in a syscall when its ssh session closes (stdin watchdog in POSIX sh)

A child blocked in a read (e.g. on NFS) never sees stdin EOF, so closing ssh leaves it running. Fix: in the remote sh, exec 9<&0, then run the program in the background reading a fifo, and run a background watchdog that does cat <&9 >fifo. When cat ends (EOF), wait a few seconds, then for each descendant (pgrep -P, recursive) send kill -STOP and then kill -KILL, children first, so the parent cannot respawn workers. Only kill pids whose ppid (ps -o ppid= -p) is still your shell, so a reused pid is never hit. Gotchas: (1) in non-interactive sh, an async list gets stdin from /dev/null, so dup stdin to fd 9 before backgrounding; (2) close the fifo write end in every other process, or EOF never arrives; (3) the remote shell and the watchdog need trap : HUP TERM, because a signal to the process group (Foundation Process.terminate() on macOS reached the whole local group in tests) otherwise kills the watchdog and leaves the stuck child; (4) keep the top shell waiting on the program and exit with its status, so the caller still sees the exit code.