### What happened?

Two separate defects combine to make every Azure gpt-5.6 request carrying a `temperature` other than `1` fail with an Azure 400, with no supported way to work around it through `model_info`

``` AzureException BadRequestError - Unsupported value: 'temperature' does not support 0.0 with this model. Only the default (1) value is supported. ```

**Defect A: per-deployment capability overrides never reach the gpt-5 parameter gates**

Router registration puts a deployment's `model_info` into `litellm.model_cost` under the backend deployment name (`azure/<deployment>`), but the Azure branch of `get_optional_params` collapses model identity to `_azure_detection_model = base_model or model` (`litellm/utils.py:4287`) and passes only that into `AzureOpenAIGPT5Config.map_openai_params`. Every capability gate in `litellm/llms/openai/chat/gpt_5_transformation.py` then resolves `supports_none_reasoning_effort` against `base_model`, so a per-deployment override is silently ignored

Setting `supports_none_reasoning_effort: false` on the deployment has no effect: `/model/info` reports the flag as `False` while the temperature is still forwarded to Azure. This affects the tempera…