fix(examples): guard against empty choices in LLM completion response

[qizwiz]

What

Guards completion.choices[0] in examples/demo/agents/openai_completion_agent.py before accessing .text:

# Before
action = util.extract_json(completion.choices[0].text, ["/END"])

# After
if not completion.choices:
    raise ValueError("LLM returned empty choices list")
action = util.extract_json(completion.choices[0].text, ["/END"])

Why

The OpenAI Completions API can return choices=[] at HTTP 200 when content is filtered. Without the guard, this raises an unhandled IndexError that crashes the agent mid-action, losing any state built up during the conversation.

Evidence

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

Scope

1 file, 2 lines added.