cudaPackages: fix multiplex builder isSupported check

[ConnorBaker]

Description of changes

The predicate was incorrect because package.platform is the name of an NVIDIA redistArch and so will always be correct (the releases.nix expression is based on the manifests NVIDIA provides). Instead we should check that package.platform is the same as the one we're trying to build for.

Things done

• Built on platform(s)
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• For non-Linux: Is sandboxing enabled in nix.conf? (See Nix manual)
  • sandbox = relaxed
  • sandbox = true
• Tested, as applicable:
  • NixOS test(s) (look inside nixos/tests)
  • and/or package tests
  • or, for functions and "core" functionality, tests in lib/tests or pkgs/test
  • made sure NixOS tests are linked to the relevant packages
• Tested compilation of all packages that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage
• Tested basic functionality of all binary files (usually in ./result/bin/)
• 24.05 Release Notes (or backporting 23.05 and 23.11 Release notes)
  • (Package updates) Added a release notes entry if the change is major or breaking
  • (Module updates) Added a release notes entry if the change is significant
  • (Module addition) Added a release notes entry if adding a new NixOS module
• Fits CONTRIBUTING.md.

---

Add a 👍 reaction to pull requests you find important.

[SomeoneSerge]

❯ nix eval -I nixpkgs=flake:github:ConnorBaker/nixpkgs/fix/cudaPackages-multiplex-builder-isSupported-check -f '<nixpkgs>' --arg config '{ allowUnfree = true; cudaSupport = true; }' \
  cudaPackagesGoogle.cudnn_8_6.brokenConditions
{ "CUDA version is too new" = true; "CUDA version is too old" = false; }

...expected: all false, tensorflow.outPath evaluat-able without allowBroken

Cf. #281568

[SomeoneSerge]

I guess the platform regression is independent of the cuda release regression

[GaetanLepage]

Result of nixpkgs-review pr 281806 run on x86_64-linux 1

[SomeoneSerge]

Yes I see the regression.

❯ nix eval -f https://github.com/NixOS/nixpkgs/archive/a33557d1b95a8ec416c9c047723b5bb179b57d99.zip --argstr system aarch64-linux --arg config '{ cudaCapabilities = [ "6.2" ]; cudaEnableForwardCompat = false; cudaSupport = true; allowUnfree = true; }' -L cudaPackages.cudnn.src.url
"https://developer.download.nvidia.com/compute/cudnn/redist/cudnn/linux-x86_64/cudnn-linux-x86_64-8.9.7.29_cuda12-archive.tar.xz"
❯ nix eval -f . --argstr system aarch64-linux --arg config '{ cudaCapabilities = [ "6.2" ]; cudaEnableForwardCompat = false; cudaSupport = true; allowUnfree = true; }' -L cudaPackages.cudnn.src.url
"https://developer.download.nvidia.com/compute/cudnn/redist/cudnn/linux-aarch64/cudnn-linux-aarch64-8.9.5.30_cuda12-archive.tar.xz"

Probably introduced by the same commit where I added builtins.sort preferable .... This is horrifying, we still need something much much much simpler in place of our selection logic, even after the recent refactors...