### Check for existing issues

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

### What happened?

When using the `chatgpt` provider with `gpt-5.3-codex-spark` via the Responses API in streaming mode, function/tool call arguments are dropped.

The raw SSE stream includes the correct arguments in a `response.function_call_arguments.done` event, but LiteLLM’s Responses→Chat streaming transformation only handles `response.function_call_arguments.delta`. Since Spark emits the completed args only in `.done`, the final reconstructed `ModelResponse` contains `arguments='{}'`.

### Raw SSE evidence

Spark returns the arguments correctly here:

```text event: response.output_item.added data: {"type":"response.output_item.added","item":{"type":"function_call","arguments":"","call_id":"call_...","name":"simple_add"}}

event: response.function_call_arguments.done data: {"type":"response.function_call_arguments.done","arguments":"{\"a\":5,\"b\":7}","output_index":1}

event: response.output_item.done data: {"type":"response.output_item.done","item":{"type":"function_call","arguments":"{\"a\":5,\"b\":7}","call_id":"call_...","name":"simple_add"}} ```

Lite…