Set up Cursor Cloud dev environment + AGENTS.md notes

[xre217]

Summary

Sets up the development environment for Grok Forge (Next.js 16 local AI studio) on Cursor Cloud and documents non-obvious caveats for future agents.

• Adds a ## Cursor Cloud specific instructions section to AGENTS.md covering how to run/verify the app, the Ollama requirement for chat, and a critical AMX-backend crash workaround.
• Update script set to npm install (Ollama, the llama3.2:3b model, and the AMX fix persist in the VM snapshot).

Environment setup performed

• npm install (Node 22).
• Installed Ollama + pulled llama3.2:3b, started ollama serve manually (no systemd in VM).
• AMX crash fix: disabled the sapphirerapids ggml CPU backend (moved libggml-cpu-sapphirerapids.so to /usr/local/lib/ollama/_disabled/) — AMX is advertised but unusable in the guest and caused llama-server to segfault on model load.

Verification

• ✅ npm run lint (runs; surfaces pre-existing react-hooks errors in committed source — not env issues)
• ✅ npm run build (compiles, all 13 routes)
• ✅ npm run dev:local + npm run verify (all 7 health checks pass)
• ✅ Hello-world: sent a chat message in /studio and received a local Ollama-powered reply

grok_forge_studio_chat_demo.mp4

Studio chat reply

Note: only AGENTS.md is changed in this PR. package-lock.json churn from npm install was reverted to keep the diff focused.

_{To show artifacts inline, enable in settings.}