Added CUDA version detect

[ArrogantGao]

I added a CUDA version detect in

function __init__()
    @assert CUDA.driver_version() >= v"11.4" "Error: CUDA.driver_version < v11.4"
    @assert CUDA.driver_version() <= v"12.1" "Error: CUDA.driver_version > v12.1"
    return nothing
end

to make sure the driver version satisfies v"11.4" <= CUDA.driver_version() <= v"12.1".

I updated JuliaPackaging/Yggdrasil#7541 so that the binary package will support the CUDA versions above, but I did not successfully support CUDA v12.2, because one of the deps we are using now only support v12.1 (https://github.com/JuliaBinaryWrappers/CUDA_full_jll.jl).

[ArrogantGao]

If I require the version to be smaller than v"12.1" the CI will fail, so I changed that to v"12.2" and the CI passed.
Does that mean the binary built with v"12.1" is actually able to work will v"12.2"?

[codecov]

Codecov Report

Merging #24 (a2678fc) into main (f635981) will increase coverage by 0.64%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main      #24      +/-   ##
==========================================
+ Coverage   95.65%   96.29%   +0.64%     
==========================================
  Files           2        2              
  Lines          23       27       +4     
==========================================
+ Hits           22       26       +4     
  Misses          1        1              

Files	Coverage Δ
src/CuTropicalGEMM.jl	100.00% <100.00%> (ø)

[GiggleLiu]

Interesting, but the problem is if you set the CUDA version to 12.2, the binary will not be downloaded.

[ArrogantGao]

I asked in slack and they told me that CUDA_full_jll is no longer supported and now we should use CUDA_SDK_jll.
I made that work and changed our binary to that, which support v12.2 directly.

[GiggleLiu]

Nice!

[ArrogantGao]

After this PR (JuliaPackaging/Yggdrasil#7541) is merged, this problem #23 should able to be fixed.