[Feature request] Tool calling support

[binarybana]

First time user of llm with the mlx backend. Really amazing UX and workflow you've got working here. Thank you!

I've downloaded several models that I'd expect to support tool calling (eg Qwen3) but it looks like tool calling is not supported with any of them, so I'm assuming it's a limitation of the plugin to expose it?

❯ llm models --options
...
MlxModel: mlx-community/Mistral-Small-24B-Instruct-2501-4bit (aliases: ms)
  Options:
    max_tokens: int
    unlimited: boolean
    temperature: float
    top_p: float
    min_p: float
    min_tokens_to_keep: int
    seed: int
  Features:
  - streaming
MlxModel: mlx-community/Qwen3-30B-A3B-4bit-DWQ-053125 (aliases: moe)
  Options:
    max_tokens: int
    unlimited: boolean
    temperature: float
    top_p: float
    min_p: float
    min_tokens_to_keep: int
    seed: int
  Features:
  - streaming
MlxModel: mlx-community/Qwen3-1.7B-4bit (aliases: smol)
  Options:
    max_tokens: int
    unlimited: boolean
    temperature: float
    top_p: float
    min_p: float
    min_tokens_to_keep: int
    seed: int
  Features:
  - streaming

Expected to see: tools in Features list above.