[codex] Add CUDA MPPI diagnostics and benchmark scenarios

[rsasaki0109]

Summary

• Add CUDA MPPI per-cycle diagnostics for solve time, rollout costs, valid rollout counts, and retreat state, with periodic ROS logging and optional CSV output.
• Expose the new MPPI diagnostics through the Python binding info dictionary.
• Add double_gap and moving_crossing controller benchmark scenarios, plus quick/cpu_gpu benchmark presets for faster local checks.

Impact

This makes Nav2 CUDA MPPI tuning easier to diagnose and makes benchmark iteration faster when adding new scenarios. The new moving-crossing benchmark gives a lightweight dynamic-obstacle smoke case without requiring a full simulator.

Validation

• colcon build --packages-select cuda_mppi_controller --cmake-args -DCMAKE_BUILD_TYPE=Release
• ros2 run cuda_mppi_controller parameter_validation_test
• ros2 run cuda_mppi_controller mppi_gpu_standalone 2048 diff
• python -m pytest python/tests -q in an isolated venv: 8 passed, 1 skipped
• ros2 run cuda_mppi_controller controller_benchmark /tmp/cuda_mppi_double_gap_quick double_gap quick
• ros2 run cuda_mppi_controller controller_benchmark /tmp/cuda_mppi_double_gap_cpu_gpu double_gap cpu_gpu
• ros2 run cuda_mppi_controller controller_benchmark /tmp/cuda_mppi_moving_crossing_quick moving_crossing quick
• ros2 run cuda_mppi_controller controller_benchmark /tmp/cuda_mppi_moving_crossing_cpu_gpu moving_crossing cpu_gpu