Support for LM Studio?

Hi @jlevy ,

pretty cool work, thanks a lot! This already pretty amazing. I had a look at warp, opencode-ai (the original one) and crush before and while they are pretty amazing as well, to me they feel too encapsulated.

I am missing a feature to use self-hosted LLMs, though. From what I can tell from the code, this is currently not supported, I did not over look something. I started to experiment on a [fork](https://github.com/jlevy/kash/compare/main...GiugAles:kash:feat/lmstudio) and it seems, that by leveraging litellm, that you already use, it is almost there. For now, I will proceed with my "proof of concept" but I am curious if you would consider integrating such a feature.

Configuration would look as follows:

```
LM_STUDIO_MODEL="my-remote-model"
LM_STUDIO_API_KEY="sk-1234"
# This is where litellm proxy is running. Alternative, LM Studio without litellm is also working.
LM_STUDIO_API_BASE="http://:4000/v1"

```

Kind regards,
Alessandro