libtorch-bin: move cudaSupport logic to package

Follow the same pattern as the majority of packages using cuda

pkgs/development/libraries/science/math/libtorch/bin.nix

@@ -4,12 +4,13 @@
 , lib
 , libcxx
 , llvmPackages
+, config
 
 , addOpenGLRunpath
 , patchelf
 , fixDarwinDylibNames
 
-, cudaSupport
+, cudaSupport ? config.cudaSupport
 }:
 
 let

pkgs/top-level/all-packages.nix

@@ -4256,9 +4256,7 @@ with pkgs;
 
   libtensorflow = python3.pkgs.tensorflow.libtensorflow;
 
-  libtorch-bin = callPackage ../development/libraries/science/math/libtorch/bin.nix {
-    inherit (config) cudaSupport;
-  };
+  libtorch-bin = callPackage ../development/libraries/science/math/libtorch/bin.nix { };
 
   tensorflow-lite = callPackage ../development/libraries/science/math/tensorflow-lite { };