[cmake] switch to FindCUDAToolkit

[characat0]

Cmake 3.17 introduced FindCUDAToolkit and deprecated FindCUDA, when cross compiling on linux, we found some weird behaviour, like having to run two or even three configure steps (JuliaPackaging/Yggdrasil#8593 with LightGBM for Julia) (JuliaPackaging/Yggdrasil#4554 (comment), Torch had the same issue).
Switching to FindCUDAToolkit should make this easier.

[jameslamb]

Thank you so much for this! I strongly support this change.

As it says in https://cmake.org/cmake/help/latest/module/FindCUDA.html

To find and use the CUDA toolkit libraries manually, use the FindCUDAToolkit module instead [of FindCUDA]. It works regardless of the CUDA language being enabled.

[jameslamb]

Double-checked the build logs and this looks like it's working great.

...
-- The CUDA compiler identification is NVIDIA 11.8.89
-- Detecting CUDA compiler ABI info
-- Detecting CUDA compiler ABI info - done
-- Check for working CUDA compiler: /usr/local/cuda/bin/nvcc - skipped
-- Detecting CUDA compile features
-- Detecting CUDA compile features - done
...
-- Found CUDAToolkit: /usr/local/cuda/include (found suitable version "11.8.89", minimum required is "11.0") 
...
[2/74] Building CUDA object CMakeFiles/histo_16_64_256-fulldata_sp_const.dir/src/treelearner/kernels/histogram_16_64_256.cu.o
[3/74] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/prediction_early_stop.cpp.o
[4/74] Building CUDA object CMakeFiles/histo_16_64_256_sp.dir/src/treelearner/kernels/histogram_16_64_256.cu.o
[5/74] Building CUDA object CMakeFiles/histo_16_64_256_sp_const.dir/src/treelearner/kernels/histogram_16_64_256.cu.o
[6/74] Building CUDA object CMakeFiles/histo_16_64_256-fulldata_sp.dir/src/treelearner/kernels/histogram_16_64_256.cu.o
...

(build link)

Based on that, all the other passing CI, and my own understanding of FindCUDA() vs. FindCUDAToolkit(), I'm confident merging this.

@shiyu1994 if you come back to this and disagree please let me know and we can change / revert it before the release (#6439).

[jameslamb]

@characat0 thanks again, come back and contribute any time!

[characat0]

Thanks for remembering @jameslamb, always a pleasure

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed.
To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.