*-sys crate design

[termoshtt]

FFI crate separation, e.g. cuda-sys, cudart-sys, and cublas-sys has benefit

• to separately link to libcuda.so, libcudart.so, libcublas.so
• to better support cargo feature gates disccused in Cargo features for CUDA version (updated Dec. 10) #5

How to split?

• cuda-sys
  • FFI for CUDA Driver API (libcuda.so)
• cudart-sys
  • FFI for CUDA Runtime API (libcudart.so)
• cublas-sys
  • FFI for cuBLAS

and so on. (edit: Drop cuFFT)

How to manage?

• each *-sys crates are managed as a subdirectory in this repository

cuda-sys/cuda-sys/Cargo.toml
         cudart-sys/Cargo.toml
         cublas-sys/Cargo.toml

[LutzCle]

Would it make sense to reject #4 and open separate PRs for each submodule? Then we could at least get cuda-sys and cudart-sys merged quickly and follow up with the rest. As @kngwyu noted in #4, it's easy to miss some parts that then hold up the entire PR.

[termoshtt]

Would it make sense to reject #4 and open separate PRs for each submodule?

Yes, I'm considering to split code generation part from #4 to enable static code generation at https://github.com/rust-cuda/cuda-sys/tree/code_generator branch (I'm going to create PR for it).
It replaces CUDA8-based binding to CUDA10-based binding which I generated on my env.

Once code generator is splitted, *-sys separation will be straight forward.