fix: don't rely on nixpkgs models-dev

[gigamonster256]

removes reliance on downstream nixpkgs packaging of models-dev

fixes darwin intel builds where nixpkgs models-dev is marked as broken for some reason

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

No duplicate PRs found

[rekram1-node]

Do we really wanna rely on the test fixture here? Can we just copy the latest file and put it in the nix directory? I think linking to the test fixture is fairly unexpected and could easily break in the future

[gigamonster256]

sure, that makes sense. I'll take a look at how the MODELS_DEV_API_JSON is used at build time, it may be able to completely ignore it/provide a minimal build version and not even need to keep a dummy copy around

[gigamonster256]

looks like models.json at runtime is read from 1. host cache, 2. build time provided/fetched 3. runtime fetch

since build time fetch is disallowed in nix without specifying the hash, the only long term solution not requiring continual manual intervention would be relying on nixpkgs packaged version of models.dev

the only reason we got to this state of failing builds is because previous CI unconditionally updated nixpkgs even of builds failed

i'll open a PR to revert the nixpkgs bump and set up CI for reviewed merge of nixpkgs updates