fix: Add a working cuda feature

[winston-h-zhang]

Resolves lurk-lang/arecibo#259 downstream. This PR adds an actual cuda feature that can be toggled grumpkin-msm, and thus all of its dependents. We also add GPU CI checks.

[samuelburnham]

Nit: we're only building here, not testing. Is the latter intended for later or not a priority?

[huitseeker]

1. do we really need to push the semantics of the cuda feature inside grumpkin-msm or can we simply not import grumpkin-msm in dependents if that feature is not activated?
2. can we make grumpkin-msm print an indiscutably spectacular message if it's loaded and does not manage to resolve to a GPU-using MSM? I really want the fallback to the CPU to be extremely noticeable, 100% of the time, every time.

[huitseeker]

To reproduce the comment from lurk-lang/arecibo#259, this is going in a good direction given the fresh benchmarks results, but:

we should make grumpkin-msm:
1/ always compile the CPU method, as it does now,
2/ only attempt to compile and run the GPU method when the cuda feature is enabled
3/ emit an indiscutably spectacular message when it's compiled with the cuda feature and does not find the CPU,

[winston-h-zhang]

This is the error message:

paperspace@psrjp5jvivu1:~/lurk-lab/grumpkin-msm$ cargo build --features cuda
   Compiling sppark v0.1.5
   Compiling grumpkin-msm v0.1.0 (/home/paperspace/lurk-lab/grumpkin-msm)
warning: [email protected]: feature "cuda" was enabled but no valid installation of CUDA was found
warning: [email protected]: the crate's default CPU methods will be compiled instead; NO GPU IMPLEMENTATION WILL BE USED WHEN CALLING THIS CRATE'S METHODS
warning: [email protected]: please recompile without feature "cuda" or provide a valid CUDA installation
    Finished dev [unoptimized + debuginfo] target(s) in 2.48s

[huitseeker]

Excellent, thanks a lot!