Add reasoning effort 'none' for GPT 5.1 and onward

[jpaarhuis]

Add none reasoning effort level for GPT-5.1 and newer models

Summary

This PR adds support for the none value in the reasoning_effort parameter, which was introduced for GPT-5.1 and newer models.

Changes

• Added "none" to the ReasoningEffort union in the TypeSpec specification
• Updated documentation to reflect that this applies to "Reasoning models" (not just o-series)
• Regenerated code to include ChatReasoningEffortLevel.None and ResponseReasoningEffortLevel.None

Background

According to the OpenAI API documentation, the reasoning_effort parameter now supports the following values: none, minimal, low, medium, and high.

Usage

var options = new ChatCompletionOptions
{
    ReasoningEffortLevel = ChatReasoningEffortLevel.None
};

[jsquire]

Hi @jpaarhuis. Thanks very much for your contribution and for your interest in improving the OpenAI developer experience.

Your implementation looks good, with the caveat that the TSP is owned and published upstream. We've been making minor adjustments like this by hand but are about to adopt a very large spec rewrite that is likely to overwrite this.

As a result, I'll ask @joseharriaga to make the final call on merge. Unfortunately, that means that we'll need to wait until after the US holidays.

[christothes]

Hi @jpaarhuis -
One step you'll need to complete before this is merged is to run the Export-API script to update the public API surface file.

You'll need the GenAPI tools which can be found here https://github.com/dotnet/sdk/tree/main/src/Compatibility/GenAPI

[jpaarhuis]

Ok @christothes, just ran that and committed the additional code.