Port to ROS 2 Jazzy / ament_cmake - #6

First real run on a live runner failed immediately: ROS 2's setup.bash
references AMENT_TRACE_SETUP_FILES without guaranteeing it's set, which
set -euxo pipefail's -u treats as fatal. ci_build_and_test.sh already
had this exact workaround; this inline step never got it.

None of the 4 repos' Harmonic ports are merged upstream yet, so cloning
uos/*'s main/ros2 pulled unported code and failed for reasons unrelated
to this repo's own change. Temporary, marked clearly in the workflow --
revert to uos/* before opening the PR upstream.

Unlike GitHub-hosted runners (a fresh VM every run), a self-hosted
runner's _work directory persists between runs -- plain git clone fails
once the target directory exists from a prior run. First real failure
on the newly-registered GPU runner.

The first genuine end-to-end GPU CI run (all environment/CI plumbing
issues resolved) got 13 messages in the previous 5.0s/min_messages:15
window instead of the usual 20-32 -- real timing margin on real
hardware, not a plumbing bug. Widened to 8.0s/min_messages:10 for
headroom instead of just hoping a retry passes.

- radaray_opti.py: --material-index now accepts multiple indices,
  tuning several materials' velocity/ambient/diffuse/specular in one
  differential_evolution run (was previously hardcoded to exactly one).
  Verified against a real end-to-end run with known hidden ground truth,
  not just a syntax check -- see MIGRATION_HANDOFF.md's new Phase 6 for
  the honest result (plumbing is correct, full 8-param recovery on this
  simple synthetic scene did not fully converge even at 800 evals).
- testdata/two_walls_test.dae + _materials.yaml: new synthetic 2-object
  scene built specifically to exercise/verify the above, since no other
  available scene has more than one distinctly-labeled material.
- MIGRATION.md: re-checked mesh_publisher.cpp's mesh_msgs dependency is
  still ROS 1-only upstream, not just assumed still true.
- radaray_opti.py: documented the benign, rclpy-source-confirmed cause
  of the intermittent "unexpected result response" warning.
- ARCHITECTURE.md / QUICKSTART.md / INTEGRATION_GUIDE.md: new -- how the
  4 packages fit together, hands-on runbook, and a guide for attaching
  these sensors to an existing Gazebo Harmonic sim.
- MIGRATION_HANDOFF.md: moved here from the (non-git) workspace root so
  it's actually versioned, plus the new Phase 6 writeup.

mesh_msgs and its RViz1 consumer (rviz_map_plugin/mesh_tools) are both
still ROS 1-only upstream (re-checked, not just assumed). Re-targeted
this node at visualization_msgs/MarkerArray instead -- one TRIANGLE_LIST
Marker per mesh, the same message type ray_reflection_test.cpp already
publishes successfully. RViz2 renders it natively, no external plugin
or dependency needed anywhere in the chain.

Ported the ROS side to rclcpp, kept the same map_file/pre_transform
parameter shapes, wired the executable into CMakeLists.txt next to
ray_reflection_test. Runtime-verified against testdata/two_walls_test.dae
-- ros2 topic echo /mesh_markers confirmed the published triangle points
are the mesh's actual vertices, not just "builds without crashing."

See MIGRATION_HANDOFF.md's Phase 7 and MIGRATION.md for the full writeup,
including a note on an unrelated, pre-existing SIGINT/SIGTERM shutdown
behavior shared with radar_simulator.cpp/ray_reflection_test.cpp (already
handled by the fixture harness's SIGKILL fallback, not a new issue).