OpenAI Responses message content uses the wrong completion index

[yaodong-shen]

Description

get_response_output_message_attributes() uses the index within message.content when extracting ResponseOutputText, instead of the message's index within response.output.

This mis-associates message text whenever another output item (commonly a reasoning item) precedes the message. The message metadata is emitted under gen_ai.completion.1.*, while its content is emitted under gen_ai.completion.0.content and can overwrite attributes belonging to the reasoning item.

Reproduced on current main at f8e907b92dabe47232978023fdcb01e2a7d4b752.

Minimal reproduction

from openai.types.responses import (
    ResponseOutputMessage,
    ResponseOutputText,
    ResponseReasoningItem,
)

from agentops.instrumentation.providers.openai.attributes.response import (
    get_response_output_attributes,
)

reasoning = ResponseReasoningItem(
    id="reasoning_1", summary=[], type="reasoning", status="completed"
)
message = ResponseOutputMessage(
    id="msg_1",
    content=[ResponseOutputText(annotations=[], text="answer", type="output_text")],
    role="assistant",
    status="completed",
    type="message",
)

print(get_response_output_attributes([reasoning, message]))

Relevant current output:

gen_ai.completion.0.id = reasoning_1
gen_ai.completion.0.content = answer
gen_ai.completion.1.id = msg_1
# gen_ai.completion.1.content is missing

Expected behavior

The text content should use the parent output message's completion index:

gen_ai.completion.0.id = reasoning_1
gen_ai.completion.1.id = msg_1
gen_ai.completion.1.content = answer

Suggested fix

Pass the outer response.output index through to get_response_output_text_attributes() instead of replacing it with the inner content-list index. Add a regression using actual OpenAI response types with a reasoning item before the message.

I can submit a focused fix with regression coverage.