[Provider][Regression] default NVIDIA model is not found by provider on jetson-thor

## Summary

NemoClaw v0.0.112 non-sanity `T6158298` on `jetson-thor` onboards OpenClaw with the product default NVIDIA provider/model, but the first agent inference fails with:

`GatewayClientRequestError: FailoverError: The selected model was not found by the provider. Check the model id or choose a different model.`

Internal tracking: NVBug 6646006.

## Evidence

- GitLab pipeline: `63836043`
- GitLab job: `406312286`
- Template/platform: `T6158298` / `jetson-thor`
- Observed provider/model: `nvidia-prod` / `nvidia/nemotron-3-super-120b-a12b`
- Result: template failed before CLI flag-forwarding assertions could be judged.

## Expected

The default NVIDIA model selected by onboarding should be a live provider catalog member, so first inference succeeds and CLI flag forwarding can be tested.

## Actual

The default model selected by onboarding is rejected by the provider as not found.

## Contract References

- Product source defines `DEFAULT_INFERENCE_MODEL = "nvidia/nemotron-3-super-120b-a12b"`.
- `T6146179` asserts the wizard/default cloud model must be present in the live NVIDIA endpoint catalog; otherwise accepting defaults 404s on first inference.
- `T5882196` asserts installer/onboard selects NVIDIA Endpoints and reports the required default model with the NVIDIA provider.

## Impact

This gates DevTest for `jetson-thor/T6158298` until the default model/catalog/provider configuration is corrected or the default is changed to a live catalog member.