Q1_0: port CUDA kernels

[pwilkin]

Overview

CUDA kernels for Q1_0 ported from the original fork.

Additional information

The CUDA kernels from the original fork's Q1_0_g128.

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: Yes, used Kimi 2.5 to compare the branch and port the kernels

[am17an]

I think the original authors already had planned CUDA kernels. If so, we should let them add them

[pwilkin]

@khosravipasha can you please take a look and comment?

[khosravipasha]

Oh thanks, did not see this just submitted the CUDA PR: #21629
was waiting for Metal backend to get merged.
I might need some help with tuning the kernels not a GPU expert myself, but so far the speed ups were satisfactory.

[pwilkin]

Ah, no worries :)

Obsoleted by #21629