Add Nemotron 3 Ultra to GitLab Duo Self-Hosted supported models

Adds NVIDIA Nemotron 3 Ultra to the supported models table for GitLab Duo Self-Hosted.

Related to #606732.

Please confirm before merging: the functionality levels are not defined in the evaluation issue. I matched them to comparable agentic open-weight models (Kimi, GLM, MiniMax).

Feature	Level
Code completion	Limited
Code generation	Limited
GitLab Duo Non-Agentic Chat	Partial
GitLab Duo Agent Platform	Partial

Evidence

The SWE-Bench Pro eval (30 tasks, NVFP4 via Fireworks) resolved 40.0%, completed 76.7%, and errored on 23.3%. Seven runs submitted an empty patch. That supports Partial rather than Full for Agent Platform.

The history entry will be updated with the link to the model support MR.