[codex] Fix NVIDIA hosted model slugs

[ChunkyMonkey11]

Summary

Updates the NVIDIA fast and fallback profiles to use the hosted Build/API model slugs that work on https://integrate.api.nvidia.com/v1.

Root cause

npm run claude:fast was routing claude-3-5-sonnet-latest to the Hugging Face-style BF16 model ID nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16. NVIDIA's hosted endpoint returned 404 page not found for that model ID. Direct provider probes confirmed the hosted slug nvidia/nemotron-3-super-120b-a12b succeeds on the same endpoint and account.

The Nano fallback had the same stale-slug pattern, so it was updated to nvidia/nemotron-3-nano-30b-a3b.

Changes

• Updates generated NVIDIA provider defaults and profiles.
• Updates docs and wrapper model help text.
• Updates release smoke and unit test assertions.
• Adds NVIDIA extraBody defaults for fast/fallback profiles to disable reasoning traces in normal Claude responses.

Validation

• npm run typecheck
• npm test
• npm run status
• npm run claude:fast -- -p "Reply with exactly: router-smoke" returned router-smoke