test(integration): widen gateway shutdown timeouts for sanitizer overhead

The gateway shutdown sequence (mdns stop, REST server stop, transport
teardown, plugin manager shutdown, member destructors) completes well
under a second normally, but under TSan/ASan overhead on CI runners
it can exceed the launch_ros defaults (5s sigterm_timeout / 5s
sigkill_timeout). When that happens launch_testing escalates
SIGINT -> SIGTERM -> SIGKILL and the TestShutdown.test_exit_codes
assertion reports exit code -9 even though the gateway is mid-shutdown.

The TSan run for PR #398 hit exactly this on
test_scenario_thermal_protection: the gateway logged TriggerTopicSubscriber
shutdown complete and other steps but was still cleaning up when the
SIGKILL timer fired.

Set sigterm_timeout=30 and sigkill_timeout=15 in create_gateway_node so
every integration test that goes through this factory gets the longer
windows. The bigger numbers do not slow down healthy runs since the
process exits as soon as shutdown actually finishes.

Author: bburda

src/ros2_medkit_integration_tests/ros2_medkit_test_utils/launch_helpers.py

@@ -106,6 +106,14 @@ def create_gateway_node(*, port=DEFAULT_PORT, extra_params=None, coverage=True):
         output='screen',
         parameters=[params],
         additional_env=get_coverage_env() if coverage else {},
+        # Default SIGINT->SIGTERM escalation is 5s and SIGTERM->SIGKILL is 5s.
+        # Under TSan/ASan the gateway shutdown sequence (mdns stop, REST server
+        # stop, transport teardown, plugin shutdown) easily exceeds 5s on slower
+        # CI runners, causing launch_testing to escalate to SIGKILL and the
+        # TestShutdown.test_exit_codes check to report -9. The process still
+        # shuts down cleanly given enough time, so widen both windows.
+        sigterm_timeout='30',
+        sigkill_timeout='15',
     )