chore: migrate rust/llm_chatbot to icp-cli

[marc0olo]

Summary

• Replaces dfx.json with icp.yaml; the llm canister uses a pre-built WASM (v0.3.1, Ollama mode)
• The ic environment excludes llm — icp deploy does not redeploy it on mainnet
• icp deploy auto-injects PUBLIC_CANISTER_ID:llm into the backend canister when deploying locally; ic-llm 1.2.0 reads that env var at runtime to reach the local LLM, and falls back to the hardcoded mainnet principal (w36hm-eqaaa-aaaal-qr76a-cai) when the var is absent
• Bumps ic-cdk to 0.20, ic-llm to 1.2.0, edition to 2024
• Adds backend.did (Candid interface from ic-llm ChatMessage types)
• Migrates frontend from dfx declarations + vite-plugin-environment to @icp-sdk/bindgen + actor.js using safeGetCanisterEnv
• Adds test.sh and CI workflow (rust-llm_chatbot job)
• Removes dfx.json, BUILD.md, .devcontainer/

Note for reviewers: test.sh calls prompt and chat — these require Ollama running locally (ollama serve + llama3.1:8b model pulled). CI will fail without it; Ollama support in CI can be addressed in a follow-up.

🤖 Generated with Claude Code

[marc0olo]

Blocked on ic-llm 1.2.0 being published to crates.io. That release adds default_llm_canister() which reads the PUBLIC_CANISTER_ID:llm env var that icp deploy auto-injects — without it the backend can't reach the locally-deployed LLM canister. Coordinating with the ic-llm maintainer to get the release cut.