MLX models unable to be selected for local inference

I'm trying to use `unsloth/Qwen3.6-27B-MLX-8bit` for local inference but I'm unable to select it as it reports `MLX safetensors snapshot (MLX support was not compiled in)` . If I look at the recent releases, it seems MLX model support is present. I was unable to locate any instructions on how to enable this support in desktop app.

**Expected behavior**
User should be able to use models built for MLX

---

**Screenshots**

---

**Please provide the following information**
- **OS & Arch:** macOS (Apple Silicon) 26.6 (M5 Max)
- **Interface:** UI (Desktop App)
- **Version:** v1.45.0
- **Extensions enabled:** All (Investigation focused on core server process)
- **Provider & Model:** Local inference: unsloth/Qwen3.6-27B-MLX-8bit
---

Do not begin implementation until the issue reaches **Ready** on the [Goose Issues board](https://github.com/orgs/aaif-goose/projects/1).

The issue indicates MLX support was not compiled into the desktop app. Look for build configuration flags or features related to MLX in the Rust codebase, likely in Cargo.toml or build scripts. Check the server process that handles local model loading. Determine how model support is enabled at compile-time and what needs to be added for MLX. Verify by building with MLX enabled and testing with the unsloth/Qwen3.6-27B-MLX-8bit model.

Do mô hình lập chỉ mục viết ra từ nội dung của issue.