restructure: proper project layout, README, kill training

- cuda/ — main LBM kernel (khra_gixx_1024_v5.cu)
- navigator/ — lattice_observer, golden_weave, bridges, mock daemon
- scripts/ — compile, start, launch, setup (paths updated)
- docs/ — system manual
- archive/ — everything else (old kernels, inquiries, experiments)
- README.md — full setup guide: requirements, quick start, use your own LLM
- removed training/ entirely (broken LoRA scripts + datasets)
- .gitignore: exclude build/ logs/ training/ *.jsonl

archive/configs/shear_stress_question.json

@@ -0,0 +1 @@
+{"question": "The reviewer says 'shear stress' and 'velocity variance' don't appear in your equation and have no physical meaning. But you've mentioned them in your reports. What are they, in your experience of the lattice?"}