Support function calling

Hey there! First off, amazing initiative - this is going to add so much value for people wanting to use local and client-side LLMs. There's a similar project called [webllm](https://webllm.mlc.ai/docs/) that offers an openai-style approach to client-side LLMs, but it falls short on function calling capabilities.
I think it would be awesome if you could add function calling support to this module, at least for models that already support it (like `phi-4-mini`). Totally understand this isn't an easy task though (it is not :) ), so no pressure - feel free to close this issue if it's not something you want to tackle right now.

Keep up the great work! 👍

The issue mentions adding function calling support for models like phi-4-mini, referencing the webllm project. Start by examining the existing OpenAI-compliant API structure in the codebase, then look at how other libraries implement function calling. Check the model loading and inference logic to see where function definitions and calls could be integrated. Determine what 'done' looks like by verifying that a model can accept function definitions and return appropriate calls in its responses.

Scritto dal modello di indicizzazione a partire dal testo della issue.