Add NVIDIA NIM as a provider (OpenAI-compatible)

[AinaLuc]

Summary

• Registers NVIDIA NIM as an OpenAI-compatible provider (_build_nvidia in coworker/providers/registry.py), resolving the key from the profile or NVIDIA_API_KEY, defaulting base_url to https://integrate.api.nvidia.com/v1, following the same pattern as the other OpenAI-compatible vendors (Z AI, DeepSeek, Kimi, MiniMax, Qwen, xAI, Mistral).
• Adds a curated model label for nvidia:z-ai/glm-5.2 in coworker/providers/matrix.py.
• Adds the NVIDIA logo/provider entry to the GUI provider list and e2e fixtures.
• Fixes a loading-state bug where an empty-but-loaded model list showed the "No model connected" chip instead of the model dropdown (App.tsx / Composer.tsx), with a regression test.

Closes #364.

Test plan

• npm run build in surfaces/gui — passed
• npx vitest run src/components/Composer.voice.test.tsx in surfaces/gui — passed
• Backend pytest — could not run locally (pinned aisuite git dependency needs GitHub access unavailable in this environment); please run in CI