Ideas from rust-cuda crate

[juntyr]

For the past 3+ years, I’ve developed my own rust-cuda (https://github.com/juntyr/rust-cuda) crate with the following goals:

• support generic kernels (initially via a linker wrapper, now via macro-driven instantiation)
• write safe kernels that are safe to call

Perhaps some ideas could be useful as community work now seems to converge in this org :)