🐛 Bugfix: Enhance OpenAIModel error handling and chunk processing

[Zhi-a]

• Added validation for API response types to raise ValueError for unexpected string or dictionary responses.
• Implemented safety checks to skip non-standard chunks that lack expected attributes, logging warnings for such cases.
• Introduced unit tests to cover new error handling scenarios and ensure robust processing of API responses.

[Copilot]

Pull request overview

This PR strengthens OpenAIModel.__call__ robustness by validating unexpected API response types and tolerating non-standard streaming chunks, with new unit tests covering these scenarios.

Changes:

• Add early validation to raise ValueError when the completion API returns a str or dict instead of a stream/iterator.
• Skip and warn on streamed “chunks” that don’t expose the expected choices attribute, continuing processing for remaining chunks.
• Add unit tests for the new response-type validation and non-standard chunk handling.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
sdk/nexent/core/models/openai_llm.py	Adds response-type validation and non-standard chunk skipping/warning during streaming.
test/sdk/core/models/test_openai_llm.py	Adds tests covering string/dict API responses and streaming chunks missing choices / with empty choices.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Non-standard chunks (no choices) are appended to chunk_list. Later the code assumes the last item has a .usage attribute (chunk_list[-1].usage), which will raise AttributeError if the stream ends with a non-standard chunk (or any chunk lacking .usage). Consider either (a) not appending non-standard chunks to chunk_list, or (b) updating the usage-extraction logic to find the last chunk that actually has a non-None .usage attribute before reading token counts.

Suggested change

	# Log warning and continue processing
	if hasattr(chunk, '__str__'):
	chunk_str = str(chunk)
	logger.warning(f"Received non-standard chunk (no 'choices'): {chunk_str[:200]}")
	chunk_list.append(chunk)
	# Log warning and continue processing, but do not add unsupported
	# objects to chunk_list because later logic expects standard
	# completion chunks with the usual attributes.
	if hasattr(chunk, '__str__'):
	chunk_str = str(chunk)
	logger.warning(f"Received non-standard chunk (no 'choices'): {chunk_str[:200]}")

[Copilot]

caplog is included as a fixture argument but not used, and the test docstring mentions a warning. Either assert that the expected warning was logged (via caplog) or remove the unused fixture argument to keep the test intent clear.

Suggested change

	def test_call_chunk_without_choices_attribute_continues_processing(openai_model_instance, caplog):
	def test_call_chunk_without_choices_attribute_continues_processing(openai_model_instance):