Fix OpenAI structured chat response handling

[yinli-systems]

Summary

• replace the undefined handle_response call in OpenAIModelAdapter.structured_chat with the existing response-handling pattern used by chat
• preserve incomplete-response checking for structured completions
• add a focused unit test using a fake Azure client, without live credentials
• explicitly verify the completeness check receives the first choice and original response

Fixes #36.

Validation

Validated after merging the current upstream main:

uv run --python 3.12 \
  --with pytest --with pytest-asyncio \
  --with pydantic --with pydantic-settings --with python-dotenv \
  --with i-dot-ai-utilities --with openai --with google-genai \
  pytest tests/test_azure_openai_adapter.py -q

uv run --python 3.12 --with ruff \
  ruff check common/llm/adapters/azure_openai.py tests/test_azure_openai_adapter.py

uv run --python 3.12 --with ruff \
  ruff format --check common/llm/adapters/azure_openai.py tests/test_azure_openai_adapter.py

uv run --python 3.12 \
  --with pydantic --with pydantic-settings --with python-dotenv \
  --with i-dot-ai-utilities --with openai --with google-genai \
  python -m py_compile common/llm/adapters/azure_openai.py tests/test_azure_openai_adapter.py

Results: target test passed; Ruff lint and format checks passed; Python compilation passed.

Fault-injecting the original self.handle_response(response) call made the focused test fail with the expected AttributeError, confirming the test directly covers the reported runtime bug. Restoring the fix returned the suite to green.

AI assistance

AI assistance was used to merge current upstream changes, strengthen the response-completeness assertion, run fault-injection validation, and refine this PR description. The resulting changes and validation evidence were reviewed before submission.