[CI] Fix Nemotron nightly mixed precision checkpoints test

[b8zhong]

It's broken by #18005, which will break when the quantization is mixed precision (so all Nemotron NVFP4 checkpoints are not working). Also, I think this code should be moved to quark quantization only, and it's only a log.

---

CI States

Latest PR Test (Base): ❌ Run #26980331986
Latest PR Test (Extra): ❌ Run #26980331868

[b8zhong]

/rerun-test test_nvidia_nemotron_3_super_nvfp4.py test_nvidia_nemotron_3_super_nightly.py

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[github-actions]

Results for /rerun-test test_nvidia_nemotron_3_super_nvfp4.py test_nvidia_nemotron_3_super_nightly.py:

🚀 4-gpu-b200 (1 test): ✅ View workflow run

cd test/ && python3 registered/4-gpu-models/test_nvidia_nemotron_3_super_nvfp4.py

🚀 8-gpu-h200 (1 test): ✅ View workflow run

cd test/ && python3 registered/8-gpu-models/test_nvidia_nemotron_3_super_nightly.py

[Fridge003]

@b8zhong Looks like this logic was added by AMD in this commit 293816a#diff-65c6ac2c41977f68e460f18e35053b97089631f88a9958b0796343fccee78a67.

We need to check with them whether removing them is OK

[b8zhong]

/rerun-test test_nvidia_nemotron_3_super_nvfp4.py test_nvidia_nemotron_3_super_nightly.py

[b8zhong]

@Fridge003 Yeah, let me make the fix more defensive

[github-actions]

Results for /rerun-test test_nvidia_nemotron_3_super_nvfp4.py test_nvidia_nemotron_3_super_nightly.py:

🚀 4-gpu-b200 (1 test): ✅ View workflow run

cd test/ && python3 registered/4-gpu-models/test_nvidia_nemotron_3_super_nvfp4.py

🚀 8-gpu-h200 (1 test): ✅ View workflow run

cd test/ && python3 registered/8-gpu-models/test_nvidia_nemotron_3_super_nightly.py