Support cuda 12.8.1 and SBSA wheels

[johnnynunez]

This pull request includes several updates to the build and packaging process for improved compatibility and functionality. The most important changes include updates to the GitHub Actions workflow for building wheels, modifications to the setup.py script to better handle different architectures and platforms, and minor updates to the CUDA toolkit version.

GitHub Actions Workflow Updates:

• Updated the os matrix to include ubuntu-22.04 and ubuntu-22.04-arm and added an arch matrix to support both x86_64 and aarch64 architectures. Excluded incompatible combinations to prevent build errors.
• Added a step to set the MATRIX_ARCH environment variable based on the arch matrix value.
• Updated the CUDA toolkit version used in the workflow from v0.2.19 to v0.2.23.
• Updated the hard-coded version of pytorch-triton and the installation URLs in the workflow to support the new architecture matrix.

setup.py Script Updates:

• Renamed the get_platform function to get_arch and updated it to return the architecture of the current system. Added a new get_system function to return the system name, and updated the get_platform function to use both get_system and get_arch. [1] [2]
• Added type annotations to several functions for better code clarity and type checking. [1] [2] [3]

Why this PR @tridao ?

1. Github: https://github.blog/changelog/2025-01-16-linux-arm64-hosted-runners-now-available-for-free-in-public-repositories-public-preview/
2. windows arm q2 2025: Windows ARM hosted runners for OSS [Preview] github/roadmap#1098
3. ubuntu 20.04 is deprecated from today
4. Devices: Digits, jetson thor, cuda arm laptops are coming
5. Nvidia is merging SBSA and ARM64 together
6. I add cuda 12.8.0 and arm runners on https://github.com/Jimver/cuda-toolkit/releases/tag/v0.2.21