### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

LiteLLM Proxy v1.91.0 rejects `tool_choice` for Fireworks Kimi K3 before sending the request upstream:

```text litellm.UnsupportedParamsError: fireworks_ai does not support parameters: ['tool_choice'], for model=accounts/fireworks/models/kimi-k3 ```

The affected LiteLLM model is configured with:

```text model_name: kimi-k3 litellm_params.model: fireworks_ai/accounts/fireworks/models/kimi-k3 ```

The failure is local to LiteLLM capability validation, not an upstream Fireworks rejection:

- plain K3 completion succeeds; - K3 with tools and no tool_choice succeeds; - K3 with tools and tool_choice: "auto" fails with UnsupportedParamsError; - the exact same request succeeds when allowed_openai_params: ["tool_choice"] is supplied, and Fireworks K3 returns a structured tool_calls response.

Expected behavior: LiteLLM should allow and forward tool_choice for fireworks_ai/accounts/fireworks/models/kimi-k3, without requiring callers to set allowed_openai_params.

This appears to be missing model-registry metadata. For comparison, the working Firework…