.Net: Bug: Re-enable AzureAIInferenceChatCompletionService

[RuizhangZhou]

Describe the bug
The AzureAIInferenceChatCompletionService class is currently marked as deprecated. However, for scenarios where we want to deploy open-source models (or other models that need custom handling), having a dedicated and fully supported Azure AI Inference chat completion service is crucial. The recommended alternative—using Azure.AI.Inference.ChatCompletionsClient.AsChatClient().AsChatCompletionService()—does not work as expected in my environment. In particular, there are issues with the extension methods (e.g., no overload for AsChatClient accepting a model/deployment parameter), making it impossible to correctly configure the service.

Expected behavior

It would be ideal if one of the following were addressed:

• The AzureAIInferenceChatCompletionService should be re-enabled (or at least not marked obsolete) so that it can be used directly.
• Alternatively, the recommended pathway using Azure.AI.Inference.ChatCompletionsClient should be updated or adapted to correctly support specifying a model/deployment parameter and properly wrap the client into an IChatCompletionService without the extension method issues.

Screenshots
If applicable, add screenshots to help explain your problem.

Platform

• Language: C#
• Source: Microsoft.SemanticKernel 1.38.0, Azure.AI.Inference 1.0.0-beta.3
• AI model: Llama-3.3-70B-Instruct
• IDE: VS Code
• OS: Windows

https://github.com/microsoft/semantic-kernel/blob/6d0c6148cb0bbdb81f2e7f2430e69d325652cdb6/dotnet/src/Connectors/Connectors.AzureAIInference/Services/AzureAIInferenceChatCompletionService.cs