## Problem

`CerebrasConfig.get_supported_openai_params()` returns a hand-crafted list that omits `max_retries` and `extra_headers`. Because `map_openai_params` only passes through parameters in this list, any `max_retries` value is silently discarded. The OpenAI HTTP handler at `openai.py:653` then always uses its hardcoded default of 2 retries, ignoring the caller's intent — including `max_retries=0` to disable retries entirely.

## Proposed fix

Add `"max_retries"` and `"extra_headers"` to the list in `CerebrasConfig.get_supported_openai_params()`. Both are LiteLLM-internal params consumed by the shared OpenAI HTTP handler before the request body is assembled; neither is sent to the Cerebras API as an unexpected field.

## Files

- `litellm/llms/cerebras/chat.py`