feat: add TurboQuant support

[the-wondersmith]

PR addresses #296, adding TurboQuant support to mlx-engine's KV cache via turboquant-mlx

Note

I am not all that familiar with mlx-engine's internals, and as such am not 100% confident this implementation is the best / "correct" way to do it. I am more than happy to amend or refactor if any maintainer has input on a better way to do it.

Important

This PR only implements support for TurboQuant, it does not enable it by default, nor does it make it available/active in the LM Studio app. So far as I can tell, changes would need to be made in the LM Studio app codebase to "close the loop" here.

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}

[the-wondersmith]

---

I have read the CLA Document and I hereby sign the CLA

recheck

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 18c7284191

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[the-wondersmith]

I've just seen that mlx-lm/#1067 actually adds TQ on the mlx-lm side of things, which mlx-engine uses under the hood.

I'm going to leave the PR open for the moment, but it may be better to simply wait until that PR lands and refactor this to use mlx_lm.generate.make_turboquant_cache (which will exist at that point, but does not currently).