[codex] Add GPU graph-guided neural MPPI demo

[rsasaki0109]

Summary

• Add a CUDA graph-guided neural MPPI demo that compares unguided and experience-graph-guided rollout batches.
• Register the gpu_graph_guided_neural_mppi CMake target and generate a gallery GIF.
• Document the benchmark in the capability matrix, planning gallery, and headline table.

Validation

• git diff --check
• cmake --build build --target gpu_graph_guided_neural_mppi -j$(nproc)
• ./bin/gpu_graph_guided_neural_mppi
• ctest --test-dir build --output-on-failure --label-regex 'cpu|python|gpu' -j$(nproc)\n\n## Demo Result\n- 32768 rollouts x H=72 x guided/unguided batches\n- Cost: 1430.31 -> 842.35 (-41.1%)\n- Terminal error: 1.25 -> 0.15 (-87.8%)\n- Route error: 0.491 -> 0.045\n- GPU: 0.563 ms, 1320.1x vs CPU equivalent rollout evaluation