Make CUDA.jl an optional dependency

[kernelmethod]

Tracking JuliaLang/Pkg.jl#1285

CUDA.jl is a rather large dependency for this project to have for a feature that not everybody is going to want. Hopefully if/when conditional dependencies are added to Julia, we can feature-flag CUDA.jl for only the people who want GPU support.

[snu-lukemin]

 CUDA [052768ef] log:
 ├─possible versions are: 0.1.0-5.3.3 or uninstalled
 ├─restricted by compatibility requirements with ChaChaCiphers [f23b13a2] to versions: 3.8.0-3.13.1
 │ └─ChaChaCiphers [f23b13a2] log:
 │   ├─possible versions are: 0.1.0-0.2.0 or uninstalled
 │   └─restricted to versions * by an explicit requirement, leaving only versions: 0.1.0-0.2.0
 └─restricted by compatibility requirements with GPUCompiler [61eb1bfa] to versions: 5.0.0-5.3.3 or uninstalled — no versions left
   └─GPUCompiler [61eb1bfa] log:
     ├─possible versions are: 0.1.0-0.26.4 or uninstalled
     └─restricted by julia compatibility requirements to versions: 0.22.0-0.26.4 or uninstalled

It seems that the CUDA dependency makes ChaChaCiphers.jl unable to download (via Julia package manager) anymore.

[kernelmethod]

oops

Haven't been following Julia in a while, sadly, but it looks like the optional dependency feature got added to Pkg.jl. I'll see if I can get this fixed over the weekend (and probably upgrade the CUDA.jl version in the process).

[kernelmethod]

Haven't made CUDA.jl an optional dependency yet, but I've updated to CUDA.jl 5.3 so this error should be resolved once JuliaRegistrator updates to v0.3.0 in a few minutes.

[snu-lukemin]

Thanks, it works fine just as before!