feat: Add support for compute capability 12.1 (Grace Blackwell)

[multippt]

Motivation

When installing nunchaku on DGX Spark using the Grace Blackwell GB10 chip, installation would fail with the following error

AssertionError: Unsupported SM 121

Modifications

Whitelist Cuda GPU Compute Capability 12.1

Checklist

• Code is formatted using Pre-Commit hooks.
• Relevant unit tests are added in the tests directory following the guidance in Contribution Guide.
• Documentation and example scripts in examples are updated if necessary.
• Throughput/latency benchmarks and quality evaluations are included where applicable.
• For reviewers: If you're only helping merge the main branch and haven't contributed code to this PR, please remove yourself as a co-author when merging.
• Please feel free to join our Slack, Discord or WeChat to discuss your PR.

[lmxyy]

Did you verify the code is runnable after adding this flag?

[multippt]

Did you verify the code is runnable after adding this flag?

Yes, on the DGX Spark device, validated using the test scripts and also on ComfyUI with ComfyUI-Nunchaku custom node.

Sample svdq-fp4_r128-qwen-image-edit-lightningv1.0-4steps.safetensors, render duration 16.28s (2.65s/it), there's a performance improvement versus the equivalent vanilla ComfyUI QwenImageEdit + 4-step LoRA workflow.

[kandy]

Can confirm that this fixing issue for DGX Spark

[lmxyy]

looks good to me.