fix: guard against empty choices and message=None in LLM responses (2 files)

[qizwiz]

What

Guards response.choices[0] in two flow utility files before accessing .message.content:

chat_with_pdf/flows/standard/chat_with_pdf/utils/oai.py (OAI.stream method):

+ if not response.choices or response.choices[0].message is None:
+     raise ValueError("LLM returned empty or filtered response")
  return response.choices[0].message.content

function_flows/flows/basic/llm.py (my_llm_tool function):

+ if not response.choices or response.choices[0].message is None:
+     raise ValueError("LLM returned empty or filtered response")
  return response.choices[0].message.content

Why

The OpenAI API (and Azure OpenAI) can return choices=[] or choices[0].message=None at HTTP 200 when content is filtered by safety systems. Without the guard, promptflow pipelines receive an IndexError/AttributeError that terminates the flow run without a meaningful error message.

Azure content filtering is particularly common in enterprise deployments — this is a real failure mode for promptflow users.

Evidence

• Live production crash — confirmed in the wild from the same pattern
• Found by pact across 13.5k violations in 759 repos

Scope

2 files, 4 lines added.