Enable linker script optimization for CUDA ARM wheel

[tinglvv]

follow up on https://github.com/pytorch/builder/pull/1828/files.
WAR to enable linker script optimization.

pytorch PR pytorch/pytorch#126174
tested by pytorch/pytorch#127514

cc @Aidyn-A @ptrblck @nWEIdia @atalman

[nWEIdia]

Looks great!

[nWEIdia]

@atalman please help prioritize this one

[Aidyn-A]

Thank you very much, Ting!
It appears to be working now. Here is the log before this change (https://github.com/pytorch/pytorch/actions/runs/9229148090/job/25394747216?pr=126174 ):

2024-05-24T19:53:00.1145355Z BASE_CUDA_VERSION is set to: cu124
2024-05-24T19:53:00.1146026Z + python /builder/aarch64_linux/aarch64_wheel_ci_build.py --enable-mkldnn --enable-cuda
2024-05-24T19:53:00.1789747Z Building PyTorch wheel
2024-05-24T19:53:00.3634632Z Building wheel torch-2.4.0a0+git090c907
2024-05-24T19:53:00.6837713Z -------------------------------------------------------------------------------------------------
2024-05-24T19:53:00.6838754Z |                                                                                               |
2024-05-24T19:53:00.6839526Z |            WARNING: we strongly recommend enabling linker script optimization for ARM + CUDA. |
2024-05-24T19:53:00.6840421Z |            To do so please export USE_PRIORITIZED_TEXT_FOR_LD=1                               |
2024-05-24T19:53:00.6841104Z |                                                                                               |
2024-05-24T19:53:00.6841833Z -------------------------------------------------------------------------------------------------
2024-05-24T19:53:00.7274827Z running clean
2024-05-24T19:53:00.9344867Z Building OpenBLAS
2024-05-24T19:53:00.9352913Z Cloning into 'OpenBLAS'...

And after the change (https://github.com/pytorch/pytorch/actions/runs/9230224553/job/25398013014?pr=126174):

2024-05-24T21:46:28.1244483Z BASE_CUDA_VERSION is set to: cu124
2024-05-24T21:46:28.1245149Z + python /builder/aarch64_linux/aarch64_wheel_ci_build.py --enable-mkldnn --enable-cuda
2024-05-24T21:46:28.1892538Z Building PyTorch wheel
2024-05-24T21:46:28.3741742Z Building wheel torch-2.4.0a0+git8f729af
2024-05-24T21:46:28.7345907Z running clean
2024-05-24T21:46:28.9416764Z Building OpenBLAS
2024-05-24T21:46:28.9424643Z Cloning into 'OpenBLAS'...

[tinglvv]

Thanks @Aidyn-A for pointing out the env var typo and verifying.

[tinglvv]

@pytorchbot rebase

[tinglvv]

Closing as this PR gets Manually-specified variables were not used by the project: USE_PRIORITIZED_TEXT_FOR_LD warning. Guessing we need to inject the env var in the yml file. Opening PR in pytorch/pytorch to test.

This PR was reopened (likely due to being reverted), so your approval was removed. Please request another review.