### Check for existing issues

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

### What happened?

When trying to use the openai provider agains't an OpenAI comaptible proxy with GPT-5.2 (custom `api_base`, an unknown one, not one known by LiteLLM builtin) via `openai/chat_completions/*` or by enabling `use_chat_completions_api`, LiteLLM will work when you use the responses API, but will error when using chat/completions with invalid model ID or unknown keyword argument respectively.

Worse, with newer models like GPT-5.4 that don't support chat completions the code to internally auto route them to the responses API might trigger which again bypasses the intention of this flag.

I guess the code to handle it might not exist in this path...

(It could always be appropriate to also have a different provider for such use cases, e.g. unknown generic OpenAI "compatible" providers that should only use chat completions or should use both chat completions and responses)

### Steps to Reproduce

1. Configure LiteLLM with: ```yaml - model_name: chat-gpt-5.2 litellm_params: model: openai/chat_completions/gpt-5.2 api_key: os.environ/OPENAI_API…