.Net: OpenAI Structured Outputs ADR and implementation for Option 1 and Option 2

[Dmytro Struk (dmytrostruk)]

Motivation and Context

Related: #7946
Closes: #8519

Structured Outputs is a feature in OpenAI API that ensures the model will always generate responses based on provided JSON Schema. This gives more control over model responses, allows to avoid model hallucinations and write simpler prompts without a need to be specific about response format.

This PR contains an ADR which describes several options how to enable this functionality in .NET version of Semantic Kernel and implementation for Option 1 and Option 2.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the SK Contribution Guidelines and the pre-submission formatting script raises no violations
• All unit tests pass, and I have added new tests where possible
• I didn't break anyone 😄

[jonnermut]

So does this end up implementing both options?

To us Option#1 is essential and #2 is a nice to have.

The reason for that is that we are making Json Schemas for json output dynamically at runtime.
The do happen to be based on a C# type, but we send a subset of the types properties to the LLM depending on what we want to generate.

[Dmytro Struk (dmytrostruk)]

So does this end up implementing both options?

jonnermut That's correct.

Here is an example how to use Option 1 with manually constructed JSON Schema:

semantic-kernel/dotnet/samples/Concepts/ChatCompletion/OpenAI_StructuredOutputs.cs

Line 33 in 9d28474

public async Task StructuredOutputsWithChatResponseFormatAsync()

And this is an example of Option 2 with automatically constructed JSON Schema:

semantic-kernel/dotnet/samples/Concepts/ChatCompletion/OpenAI_StructuredOutputs.cs

Line 106 in 9d28474

public async Task StructuredOutputsWithTypeInExecutionSettingsAsync()