Feature/nix support

[jake-arkinstall]

Description

This is a WIP PR to add nix support to the repo. The aim is to combat the packaging requirements around cuda and tket.

It is not ready for merging yet. For effective use there are three things remaining:

• Documentation
• Remote Caching with cachix. At current, cachix does not skip uploading packages that are already available in the upstream cuda-maintainers.cachix.org cache, so a push would be quite large and fill up precious space. I have reached out to Domen and he has told me that an update to help us on this will be out next week.
• Usage with non-nixos systems. I have successfully run this flake on an ubuntu docker image with cuda drivers loaded, but it is awkward. We need to make use of a third-party NixGL flake and wrap binaries in it, in order to have the system's cuda drivers available for nix to use. I want to improve this somewhat, otherwise it will be confusing for users.

When this is ready and merged, CI should build any changes and upload them to cachix. I will pre-populate cachix with the heavy dependencies to make this process quick and easy. When this is all in place, we can think about use the nix flake for testing in CI using a gpu-enabled machine without being concerned about the awkward dependency setup process.

Related issues

N/A

Checklist

• [x] I have run the tests on a device with GPUs.
• [ ] I have performed a self-review of my code.
• [ ] I have commented hard-to-understand parts of my code.
• [ ] I have made corresponding changes to the public API documentation.
• [x] I have added tests that prove my fix is effective or that my feature works.
• [ ] I have updated the changelog with any user-facing changes.

[jake-arkinstall]

This is still waiting on two things:

1. Cachix environment variable - when set, built binaries will be uploaded to the cache. However, we do not wish to do this until Cachix supports adding additional upstreams, otherwise we end up uploading bulky nvidia libraries to our own cache, and we have limited storage. If we could add cuda-maintainers.cachix.org to our upstreams, this would not happen. After reaching out to Domen, he mentioned that this functionality is on its way.

2. A test runner. We need GPU capabilities on a github actions runner in order for this to work. I am happy to donate my dev machine for this but we could do with something in-office.

Additionally, I am now leaning towards the idea of centralising our nix stuff into its own 'tketpkgs' repo.