Build: fix torch build on cuda-12

[caic99]

Reminder

• Have you linked an issue with this pull request?
• Have you noticed possible changes of behavior below or in the linked issue?
• Have you explained the changes of codes in core modules of ESolver, HSolver, ElecState, Hamilt, Operator or Psi? (ignore if not applicable)

Linked Issue

conda-forge/abacus-feedstock#11

What's changed?

Torch will try to detect all applicable architectures for nvcc.
However, for TorchConfig -> Caffe2Config -> FindCUDA (outdated) -> select_compute_arch, it fails to remove unsupported arch for new CUDA toolkits.
This cause error when building with CUDA 12+:
nvcc fatal : Unsupported gpu architecture 'compute_35'

This PR will re-use the detected arch at the CUDA component part for Torch.

Any changes of core modules? (ignore if not applicable)

• Example: I have added a new virtual function in the esolver base class in order to ...

[njzjz]

Why can't the test detect this problem?

[dyzheng]

Maybe the test of cuda compilation should cover different version of cuda, thanks for your contribution.

[caic99]

Why can't the test detect this problem?

The tested CUDA environment does not contain PyTorch components.