### What happened?

Per OpenAI's parameter compatibility guide, the **gpt-5.1 family** (including `gpt-5.1-mini` and `gpt-5.1-nano`) supports `temperature` when `reasoning_effort` is set to `'none'` or omitted (the API default). LiteLLM's own pre-flight check correctly applies this for plain `gpt-5.1` (verified working below) — but `gpt-5.1-mini` and `gpt-5.1-nano` 400 with the same `UnsupportedParamsError` even when `reasoning_effort` is omitted.

Root cause appears to be a registry gap in `model_prices_and_context_window.json`:

- `gpt-5.1`: `supports_none_reasoning_effort: true` ✓ (works correctly) - `gpt-5.1-mini`: **not in registry** → falls back to "not supported" default - `gpt-5.1-nano`: **not in registry** → same

Because `OpenAIGPT5Config.map_openai_params` gates the `temperature` pass-through on `self._supports_reasoning_effort_level(model, "none")`, missing-from-registry → `False` → strict rejection. Plain `gpt-5.1` is in the registry with `supports_none_reasoning_effort=true`, which is why it works.

### Reproduction

LiteLLM v1.83.10-stable, called through standard `/v1/chat/completions` (no Responses-API routing, no Azure):

**Case A — `openai/gpt-5.1` (control, in r…