[jsk_rviz_plugins][jsk_rqt_plugins] ROS2 migration - #903

…lay linear_gauge_display normal_display normal_visual open_all_tool plotter_2d_display twist_stamped_display

Migrate the rest of the plugins whose dependencies are available in ROS 2:

- overlay: overlay_image_display, overlay_menu_display,
  overlay_diagnostic_display, overlay_picker_tool
- 3D: target_visualizer_display, tf_trajectory_display,
  pictogram_display, pictogram_array_display
- tool/display: screenshot_listener_tool, quiet_interactive_marker_display,
  video_capture_display, rviz_scene_publisher
- panel: record_action, object_fit_operator, robot_command_interface,
  select_point_cloud_publish_action, tablet_controller_panel
- shared helpers: facing_visualizer, image_transport_hints_property

Generate msg/srv with rosidl_auto_generate_interfaces(). It reserves the
${PROJECT_NAME} target name, so the plugin library target is renamed to
${PROJECT_NAME}_plugins and OUTPUT_NAME keeps libjsk_rviz_plugins.so.

Link opencv_videoio explicitly since cv_bridge does not export it, which
left cv::VideoWriter undefined in video_capture_display.

Behaviour changes forced by the ROS 2 APIs:
- robot_command_interface takes a list of button ids plus per-id parameter
  groups because ROS 2 parameters cannot hold an array of structs
- ImageTransportHintsProperty returns the transport as a plain string
- overlay_picker_tool picks String/LinearGauge instead of OverlayText/PieChart

Not ported, documented in CMakeLists.txt:
- already upstream: overlay_text_display, pie_chart_display
  (rviz_2d_overlay_plugins), camera_info_display, pose_array_display
  (rviz_default_plugins)
- waiting for upstream packages: ambient_sound*, footstep_display,
  bounding_box*, torus_array, segment_array, human_skeleton_array,
  polygon_array, simple_occupancy_grid_array,
  people_position_measurement_array, tablet_view_controller,
  yes_no_button_interface
- overlay_camera_display: rviz_rendering does not expose the Ogre render
  target behind RenderPanel::getRenderWindow()

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BmCHFJ43AcU7UNwuW2Fz4a

add_rostest() has no ROS 2 counterpart, so the rviz crash tests are rewritten
as launch tests registered with add_launch_test() from launch_testing_ament_cmake.

test/rviz_config_check.py keeps checking the same three things as the ROS 1
rostest node: the rviz node shows up, it stays alive for test_duration seconds
and it ends up subscribing to the expected topics. It is a plain mixin rather
than a unittest.TestCase so that importing it does not make launch_testing run
it on its own. The rviz node is now looked up by its node name (rviz2 always
names it `rviz`) instead of the ROS 1 rviz/SendFilePath service, and the
subscription list comes from get_subscriber_names_and_types_by_node() instead
of the rosmaster system state.

Tests, mirroring the ROS 1 test names:
- test_overlay.py                  (overlay.test)
- test_linear_gauge.py             (linear_gauge_test.test)
- test_service_call_panel.py       (service_call_panel.test)
- test_robot_command_interface.py  (robot_command_interface.test)
- test_pictogram.py                (new, ROS 1 had no pictogram test)

Each also gets a post_shutdown_test asserting rviz did not crash on exit.

Supporting assets the tests need:
- launch/*.launch.py for the five samples
- config/*.rviz converted to the rviz2 format with the migrated jsk classes
- samples/{overlay,overlay_menu,piechart,pictogram}_sample.py ported to rclpy
- config/{command_samples,default_robot_command}.yaml in the ROS 2 parameter
  layout the migrated panels expect
- install config/, icons/, launch/ and the sample scripts

ROS 1 forced DISPLAY to :0.0; the tests instead skip themselves when DISPLAY is
unset so that a headless run does not report false failures. Run them under
`xvfb-run -a colcon test ...` to exercise them without a display.

Still not ported, because the plugins or scripts they cover are not migrated
yet: boundingbox, polygon_array, segment_array, contact_state_marker,
link_marker_publisher, piechart (upstream), normal/face_detector (need the
sample bag) and the test_overlay_text_interface/point_test python unit tests.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BmCHFJ43AcU7UNwuW2Fz4a

Remove the ROS 1 rostest files whose ROS 2 launch_testing counterparts are in
place, together with the sample launch files they included:

  jsk_rviz_plugins
    test/overlay.test                    -> test/test_overlay.py
    test/linear_gauge_test.test          -> test/test_linear_gauge.py
    test/robot_command_interface.test    -> test/test_robot_command_interface.py
    test/service_call_panel.test         -> test/test_service_call_panel.py
    launch/{overlay_sample,linear_gauge_sample,pictogram_sample,
            robot_command_interface_sample,service_call_panel}.launch
                                         -> the matching *.launch.py
  jsk_rqt_plugins
    test/test_rqt_plugins.test           -> test/test_rqt_plugins.py

test/piechart.test and launch/piechart_sample.launch go as well: PieChart is
provided by rviz_2d_overlay_plugins in ROS 2, so nothing in this package can
satisfy them any more.

The rostest files of the plugins that are still blocked on unmigrated upstream
packages are kept as the reference for porting them later: boundingbox,
polygon_array, segment_array, contact_state_marker, link_marker_publisher,
normal, face_detector and test_overlay_text_interface.

Also commits the removal of the commented-out catkin dependency block from
jsk_rviz_plugins/package.xml.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BmCHFJ43AcU7UNwuW2Fz4a

image_transport 7 (rolling; jazzy has 5.1.8 and lyrical 6.4.10) changed both free
functions this package uses:

- the first parameter is a rclcpp node interfaces object instead of a
  rclcpp::Node*. It is built from a node reference, so the node is dereferenced
  at the call site
- custom_qos became a required rclcpp::QoS with no default value

There is no single expression that satisfies both signatures: passing *node_ and
rclcpp::QoS(10) unconditionally fails on jazzy with "cannot convert
'rclcpp::Node' to 'rclcpp::Node*'", so the call sites are guarded with
IMAGE_TRANSPORT_VERSION_GTE(7, 0, 0). QoS(10) reproduces the
rmw_qos_profile_default the older signature defaulted to, which keeps the QoS
identical across distros; rclcpp::SystemDefaultsQoS() would instead leave every
policy to the RMW and change the behaviour on rolling only.

Verified by building and testing the workspace in a ros:rolling-ros-base
container with the same steps industrial_ci runs: 2 packages built,
23 tests, 0 errors, 0 failures, 14 skipped (launch tests, no DISPLAY).

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BmCHFJ43AcU7UNwuW2Fz4a

Drop the descriptions of what the ROS 1 rostest counterparts used to do; they
say nothing about the code that is here.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BmCHFJ43AcU7UNwuW2Fz4a

rclpy dropped Logger.warn() after jazzy, so label.py raised AttributeError on
lyrical and rolling.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BmCHFJ43AcU7UNwuW2Fz4a

The tests skip themselves without a display, so nothing was exercised on CI.
industrial_ci runs the BEFORE_RUN_TARGET_TEST hook in a sub shell, hence the
_EMBED variant, which is eval'd in the shell that runs the tests.

Under software OpenGL the processes corrupt the heap while tearing down their
Fast DDS subscriptions and die from SIGABRT, so the post shutdown tests accept
that. It happens on lyrical and rolling, not on jazzy.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BmCHFJ43AcU7UNwuW2Fz4a

… on exit

rqt_plot renamed ROSData.next() to next_data() after jazzy, which left the
plot and the label without data on lyrical and rolling.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BmCHFJ43AcU7UNwuW2Fz4a