debug-on-exit.sh: use 'timeout' for remote_command

I've seen CI runs where remote_command spends tens of minutes trying and never managing to run a kubectl debug node on a broken node.

This MR protect this command with a timeout 30s to avoid this.