- add the NVIDIA chat-completions provider and Nemotron 3 Super model config
- force reasoning on for NVIDIA requests and strip explicit token and reasoning budget fields
- extend provider, runtime, catalog, adapter, desktop label, and regression test coverage