autoware_ground_filter test_node times out at TestParameterUpdateBranches due to multiple "GroundFilter" nodes in the same context

[xmfcx]

Description

In CI the test_node gtest target of autoware_ground_filter times out at the
60 s CTest limit. The suite gets through 24 of 25 cases successfully, then
hangs on GroundFilterComponentTest.TestParameterUpdateBranches. The last
log line emitted before the timeout is the well-known rosout warning:

2: [ RUN      ] GroundFilterComponentTest.TestParameterUpdateBranches
2: [WARN] [...] [rcl.logging_rosout]: Publisher already registered for node name: 'GroundFilter'.
2/7 Test #2: test_node ........................***Timeout  60.04 sec

Failing CI run

• Workflow: build-and-test on main, commit a5c9fcd
• Run: https://github.com/autowarefoundation/autoware_core/actions/runs/24817457935
• Job: bnt (jazzy, false, false) / build-and-test-jazzy, step Test, line 6863

Root-cause analysis

The warning is the same one that appears in many other passing cases in this
suite. It is rclcpp telling us that a node named GroundFilter already exists
in the process and a second one is being constructed.

That is exactly what TestParameterUpdateBranches does:

1. GroundFilterComponentTest::SetUp() (test/test_node.cpp:147) calls
   rclcpp::init and constructs node_, a GroundFilterComponent named
   GroundFilter. It also creates result_sub_, a subscription on node_'s
   output topic.
2. The body of TestParameterUpdateBranches (test/test_node.cpp:1181) then
   constructs a second GroundFilterComponent with the same name
   GroundFilter, while the first one is still alive.
3. Each GroundFilterComponent constructor:
   • declares ~30 parameters,
   • creates publishers / subscribers,
   • calls setupTF() which builds a tf2_ros::TransformListener with the
     default spin_thread = true, spawning a dedicated executor thread that
     subscribes to /tf and /tf_static.
4. The test then calls node->set_parameter(...) six times in a row. With two
   GroundFilter-named nodes plus their TF-listener helper threads coexisting
   in one process, one of these set_parameter calls never returns and the
   test hangs to the 60 s timeout.

onParameter (src/node.cpp:659) takes mutex_ and calls get_parameter(...)
unconditionally for several names. If parameter machinery interacts with
duplicate-name node state (e.g. via rosout / parameter services), this is a
plausible deadlock surface.

The smoking-gun evidence that this is the trigger: every other test case that
emits the same Publisher already registered for node name: 'GroundFilter'
warning is one that also constructs a second GroundFilterComponent. The
unique thing about TestParameterUpdateBranches is that it then aggressively
mutates parameters on that second instance.

Suggested fix

Don't run a second GroundFilterComponent while the fixture's node_ is
alive. Either:

• Call node_.reset() at the top of TestParameterUpdateBranches before
  constructing the local one, or
• Reuse the fixture's node_ and call node_->set_parameter(...) on it (this
  mirrors TestParameterUpdate at test/test_node.cpp:345, which does pass).

Longer-term, two GroundFilterComponent instances should be safe to
co-exist (just give them distinct names via NodeOptions), but the immediate
fix is to stop creating duplicates in the test.

Reproduction

colcon test --packages-select autoware_ground_filter \
  --ctest-args -R '^test_node$' --output-on-failure

The hang is timing-sensitive (CI hits it; local runs may not), but the
duplicate-node pattern is deterministic in the source.