use cu126 for 10 series and older GPUs

[w-e-w]

alternate PR for

• Supported pytorch added for CUDA 11.3 and CUDA 11.7, dynamically. (Windows only) #17163
• Supported Pytorch added for Maxwell. #17171

apparently 10 series and older gpus don't work with cu128
10 series gpu have CUDA Compatibility of 7
and since GPU with Compatibility 7.2 is considered legacy by Nvidia
I belive everything under 7.2 should use cu126

ref
https://docs.nvidia.com/deeplearning/cudnn/backend/latest/reference/support-matrix.html
https://developer.nvidia.com/cuda-gpus
https://download.pytorch.org/whl/cu128
https://en.wikipedia.org/wiki/CUDA
https://discuss.pytorch.org/t/uninstall-pytorch-completely-to-install-older-version/223551
https://dev-discuss.pytorch.org/t/cuda-toolkit-version-and-architecture-support-update-maxwell-and-pascal-architecture-support-removed-in-cuda-12-8-and-12-9-builds/3128

I don't fully understand the situation with older gpus
from the information I can find it seems that they only deleted support on pytorch 2.8
and since we use 2.7 I'm not quite sure why it doesn't work with cu128

Checklist:

• I have read contributing wiki page
• I have performed a self-review of my own code
• My code follows the style guidelines
• My code passes tests

[w-e-w]

@kavyamali can you test this

should work but just to make sure

[kavyamali]

@kavyamali can you test this

should work but just to make sure

Testing right now, and it's currently downloading the correctly mentioned cu126. But it fails installation due to --no-cache-dir not being mentioned. If you want, I can edit my original code to default to cu126 for pascal as well. Or you can do so and add --no-cache-dir in your code as well. Thanks for your support. Everything else is perfect here.