Use py-cpuinfo to detect SIMD_WIDTH in platform-independent way

[lostmsu]

This enables AVX2 and AVX512 detection on Windows.

[RezaYazdaniAminabadi]

Thanks @lostmsu for the PR 👍 I have tried this on AMD and Intel CPUs and both passed the unit test.

[lostmsu]

How do I correct formatting? The test does not tell what is wrong with it.

[jeffra]

How do I correct formatting?

Can you run our pre-commit formatter? See our contributor guide here: https://github.com/microsoft/DeepSpeed/blob/master/CONTRIBUTING.md#prerequisites

[lostmsu]

@jeffra you might want to make the formatting check run without requiring approval.

[jeffra]

Thank you for the contribution @lostmsu!

[lostmsu]

@jeffra do I need to look into the test failures?

[lostmsu]

@RezaYazdaniAminabadi @jeffra I have another commit built on top of this one to support CPU Adam and Adagrad on Windows: lostmsu@4daa425

Just nagging here to get this in so I could publish another PR )

[jeffra]

@jeffra do I need to look into the test failures?

We were seeing very strange test runner failures all day yesterday, for some reason tests would be marked Abandoned after 15min or running. We've been debugging and testing to see if we could resolve the issue but today it is "magically" working again, it might have been an intermittent github actions server-side issue. If tests pass now then we can merge this today.