test(smoke): wait for opcua_bridge subscription before PLC inject

Smoke flaked on CI: inject-photoeye-flicker fired ~540 ms before the
opcua_bridge completed its AlarmConditionType subscription handshake
with plc-sim. The OPC UA raise event then had no listener, the bridge
caught only the later clear events from restore-line, and the
MANYMOVE_PLC_PHOTOEYE_FLICKER fault never made it to FaultManager.

Poll the bridge container logs for 'subscribed to AlarmConditionType
events' before the inject, plus a 1s margin for the asyncua handshake
to complete on the plc-sim side.

Author: mfaferek93

tests/smoke_test_manymove_industrial.sh

@@ -81,6 +81,29 @@ if api_get "/components"; then
     fi
 fi
 
+# Wait for opcua_bridge to actually subscribe to AlarmConditionType
+# events before injecting. Container manifest discovery (above) only
+# proves the gateway routed the component, not that the asyncua
+# subscription handshake on plc-sim completed. Without this wait the
+# inject can fire 100-500ms before bridge is ready, the OPC UA raise
+# event goes unheard and the test flakes.
+BRIDGE_LOGS=""
+for _ in $(seq 1 20); do
+    BRIDGE_LOGS=$(docker logs manymove_industrial-opcua-bridge-1 2>&1 || true)
+    if echo "$BRIDGE_LOGS" | grep -q "subscribed to AlarmConditionType events"; then
+        pass "opcua_bridge subscribed to AlarmConditionType events"
+        break
+    fi
+    sleep 1
+done
+if ! echo "$BRIDGE_LOGS" | grep -q "subscribed to AlarmConditionType events"; then
+    fail "opcua_bridge never subscribed to AlarmConditionType events"
+fi
+# Belt-and-suspenders: small extra delay so subscription is fully active
+# on the plc-sim side too (asyncua handshake completes ~50-200ms after
+# the bridge logs the subscribe call).
+sleep 1
+
 # Photoeye flicker injection -> MANYMOVE_PLC_PHOTOEYE_FLICKER (WARN)
 if curl -fsS -X POST -H "Content-Type: application/json" -d "$EXEC_BODY" \
     "$API_BASE/components/conveyor-line/scripts/inject-photoeye-flicker/executions" >/dev/null; then