Please allow us to disable reasoning for low-latency interactions

GPT 5.2 and 5.4 support the following reasoning levels:

- none
- low
- medium
- high
- xhigh

where **none** disables reasoning entirely.

However, the API doesn't allow us to use **none** - it's not exposed through `SupportedReasoningEfforts`, and attempting to use it results in fallback to the default of **medium**. Can support for **none** be added, please?

The same applies to Sonnet, for which there is no means to disable reasoning completely.

Right now, the only available model suitable for low-latency interactions is GPT 4.1.