fix: kill orphaned start-bridge.sh supervisors on restart

Prevents port 7890 conflicts caused by stale start-bridge.sh processes
surviving control-agent restarts.

The issue occurred when:
1. Manual bridge restarts left orphaned start-bridge.sh supervisors running
2. 'baudbot restart' would start a new tmux-managed bridge
3. Both supervisors competed for port 7890 → EADDRINUSE errors

This fix adds explicit cleanup of start-bridge.sh processes before
launching the new tmux-managed bridge.

Resolves recurrent bridge startup failures after manual interventions.

Author: Baudbot

pi/skills/control-agent/startup-pi.sh

@@ -94,6 +94,21 @@ mkdir -p "$BRIDGE_LOG_DIR"
 #     and any leftover old-style PID-file supervisor.
 echo "Cleaning up old bridge..."
 
+# Kill any standalone start-bridge.sh supervisors (from manual restarts or
+# deployments before tmux-managed bridge was introduced). These can survive
+# control-agent restarts and cause port conflicts with the new bridge.
+START_BRIDGE_PIDS=$(pgrep -f "start-bridge\.sh" 2>/dev/null || true)
+if [ -n "$START_BRIDGE_PIDS" ]; then
+  echo "Killing standalone start-bridge.sh supervisors: $START_BRIDGE_PIDS"
+  echo "$START_BRIDGE_PIDS" | xargs kill 2>/dev/null || true
+  sleep 1
+  # Force-kill if still alive
+  START_BRIDGE_PIDS=$(pgrep -f "start-bridge\.sh" 2>/dev/null || true)
+  if [ -n "$START_BRIDGE_PIDS" ]; then
+    echo "$START_BRIDGE_PIDS" | xargs kill -9 2>/dev/null || true
+  fi
+fi
+
 # Kill ALL tmux sessions named slack-bridge. Using list-sessions + filter
 # instead of kill-session -t handles edge cases where multiple sessions
 # somehow got the same name (e.g., from racing startups or orphaned sessions).