Feature(build): enhance CUDA version handling and architecture detection

[johnnynunez]

Support aarch64 including support GH200/GB200
Support Blackwell RTX50/A6000 and B100/B200
Support pytorch 2.7.0
Support cuda 12.8.1

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[iofu728]

Hi @johnnynunez, thanks for your help in extending support to Blackwell and ARM architectures.

However, due to two reasons: 1) We currently don't have access to the Blackwell hardware to properly verify correctness, and
2) One part of our code is still CUDA-specific and cannot be directly migrated to ARM yet,

we’re unable to immediately review and merge this PR.
That said, we truly appreciate your contribution and will revisit it once the necessary resources become available. Thanks again!

[johnnynunez]

Hi @johnnynunez, thanks for your help in extending support to Blackwell and ARM architectures.

However, due to two reasons: 1) We currently don't have access to the Blackwell hardware to properly verify correctness, and 2) One part of our code is still CUDA-specific and cannot be directly migrated to ARM yet,

we’re unable to immediately review and merge this PR. That said, we truly appreciate your contribution and will revisit it once the necessary resources become available. Thanks again!

No problems. You can check it if you want when you have access: https://pypi.jetson-ai-lab.dev/sbsa/cu128