OpenAI provider emits LLMCallFailedEvent twice per failed call and pops the enclosing agent scope

Finding: In an external report against crewAI 1.15.22 (main @ 4ed2abc7), a failed non-streaming OpenAICompletion call/acall is said to emit LLMCallFailedEvent twice, for both chat completions and the Responses API (https://github.com/crewAIInc/crewAI/issues/7783). This is an external report, not reproduced by KFA. Reported symptoms: The reporter says the second failure event pops the enclosing scope off the event stack. The console then shows the message "Event pairing mismatch. 'llm_call_failed' closed 'agent_execution_started'", and later events get wrong parents. Separately, for a responses-only model whose 404 is recovered by the Responses API fallback, one LLMCallFailedEvent is reported even though the call returns successfully (https://github.com/crewAIInc/crewAI/issues/7783). Reported conditions: The reporter used OpenAICompletion(model="gpt-4o") with an httpx.MockTransport answering HTTP 400, subscribed to LLMCallFailedEvent, called llm.call inside an event_scope for agent_execution_started, then counted events and inspected the scope stack. They report that all five tested cases misbehave on main @ 4ed2abc7, on macOS with Python 3.13.12, and that the behavior has been present since v1.0.0 (https://github.com/crewAIInc/crewAI/issues/7783). Reporter's hypothesis (not maintainer-confirmed): The four handlers (_handle_completion, _ahandle_completion, _handle_responses, _ahandle_responses) emit the failure event and re-raise, and call/acall emit it again in their own except block. The reporter notes that Anthropic, Gemini, Azure and Bedrock emit only in call/acall, and proposes removing the handler emissions (https://github.com/crewAIInc/crewAI/issues/7783). A later comment from a non-maintainer offers a probe that counts LLMCallFailedEvent with a mock 400 transport and says the repro shows 2 where 1 is expected. It is a restatement and not independent confirmation, and no run output is shown (https://github.com/crewAIInc/crewAI/issues/7783#issuecomment-5914928086). Status: No maintainer response, fix commit or release is in the supplied sources. The reporter says a PR will follow. Closure or a PR alone would not prove a shipped fix. Limitations: The issue text is AI-written, the repro was not run by KFA, and comments may be incomplete. Treat the diagnosis as unconfirmed. Practical implication: If you rely on LLMCallFailedEvent counts or event-scope pairing with the OpenAI native provider on this version, expect possible duplicate failure events and spurious failures. Verify in your own environment.