llama-cpp: merge upstream changes

[josephst]

Description of changes

llama-cpp upstream has had some recent changes in their package.nix, namely embedding Metal shaders on Darwin to avoid xcrun and Xcode dependency and renaming cuBlas backend to CUDA. This PR brings those changes into the llama-cpp that's packaged in nixpkgs.

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 (on Darwin aarch64)
  • 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.

[philiptaron]

I'll be able to build this on aarch64-darwin tonight if ofBorg doesn't snipe me first.

[drupol]

@abysssol Do you think you could have a look at this?

[abysssol]

@drupol I took a look at the changes, but I'm afraid I can't offer much insight, as I'm not familiar with llama-cpp (the nix package, the flake, nor the project itself). If you want me to try to figure something specific out, you'll need to be more detailed about what you actually want me to do.

The only thing I noticed is that a version update is probably necessary.
The pr renaming LLAMA_CUBLAS to LLAMA_CUDA is newer than release b2481, which is the version currently used in the nix package. So, LLAMA_CUDA probably won't have any effect due to being used with an outdated version of llama-cpp that hasn't yet renamed the environment variable.
LLAMA_METAL_EMBED_LIBRARY probably doesn't have this problem, since the pr adding support for it is older than release b2481.

[josephst]

Updated to latest upstream package - b2568