feat(azure): add structured output support for AzureAIChatCompletionClient

[Abhinav (abhnvgrg)]

Add structured output support for AzureAIChatCompletionClient

Closes #5957

What

Adds support for passing a Pydantic BaseModel subclass as json_output
to AzureAIChatCompletionClient.create() and create_stream(), enabling
structured output for models hosted on Azure AI Foundry and GitHub Models.

Approach

• Converts the Pydantic model to a JSON schema via model_json_schema()
  and passes it as JsonSchemaFormat(name=..., schema=..., strict=True) in
  the response_format parameter to the Azure AI Inference SDK's
  client.complete()
• Validates the response content against the Pydantic model with
  model_validate_json() after receiving the completion
• Handles both create() and create_stream() paths
• Checks model_info["structured_output"] and raises a clear error when
  the model doesn't support it
• Follows the same patterns established by OpenAIChatCompletionClient in
  PR feat: add structured output to model clients #5936

Files changed

• python/packages/autogen-ext/src/autogen_ext/models/azure/_azure_ai_client.py
  — core implementation
• python/packages/autogen-ext/src/autogen_ext/models/azure/config/__init__.py
  — type updates (if needed)
• python/packages/autogen-ext/tests/models/test_azure_ai_model_client.py
  — unit tests with mocked Azure SDK

Tests

• Structured output with create() — validates Pydantic parsing
• Structured output with create_stream() — validates streamed JSON parsing
• Model capability check — raises when structured_output is False
• Response format verification — confirms JsonSchemaFormat is passed to SDK
• All existing tests remain green

Local validation

uv run pytest packages/autogen-ext/tests/models/test_azure_ai_model_client.py -v
25 passed

uv run ruff check --select E9,F63,F7,F82 packages/autogen-ext/
All checks passed!

[Abhinav (abhnvgrg)]

@microsoft-github-policy-service agree