fix(openai): preserve async streaming response parsing

[Curtis Galione (cjgalione)]

Summary

• await asynchronous raw-response parsing before wrapping Responses API streams
• preserve the awaitable parse() contract expected by OpenAI Agents
• add a regression test that awaits and consumes the parsed async stream

Root cause

openai.AsyncAPIResponse.parse() is asynchronous. The tracing wrapper replaced it with a synchronous method while wrapping the un-awaited coroutine, so OpenAI Agents' streaming path could fail when it entered with_streaming_response and awaited api_response.parse().

The wrapper now records whether the underlying response parses asynchronously, awaits that parse before constructing the traced stream, and exposes an awaitable parse() on the returned raw-response proxy. Synchronous raw-response behavior remains unchanged.

Validation

• full OpenAI integration suites with OpenAI 2.46.0, 1.92.0, 1.77.0, and 1.71.0
• full OpenAI Agents suites with Agents 0.18.3, 0.13.6, and 0.0.19
• actual Agents 0.18.3 / OpenAI 2.31.0 streaming _fetch_response contract exercised against a mock SSE response
• all pre-commit hooks: Ruff format/check, cassette consistency, pytest-pin synchronization, and codespell
• Pylint error checks on both changed files
• final diff and customer-identifier audit