Fix the following 2 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 2
posthog/ai/stream.py:58-75
If `self._generator.aclose()` raises — for example because `_capture_streaming_event` inside the tracking generator's `finally` block throws a network error — the provider stream's `close()` is never reached. This leaves the underlying HTTP connection open until GC collects the stream, which can exhaust connection-pool slots under load. Wrapping in `try/finally` guarantees `_stream.close()` always runs regardless of what `aclose()` does.

```suggestion
    async def __aexit__(self, exc_type: Any, exc_val: Any, exc_tb: Any) -> bool:
        # Close our tracking generator first so its `finally` block runs and
        # the PostHog usage event is captured, even when the caller breaks out
        # of the loop early. If it is already exhausted this is a no-op.
        try:
            await self._generator.aclose()
        finally:
            # Then close the underlying provider stream to release the HTTP
            # connection, matching native SDK behaviour. Provider streams expose an
            # async `close()`; bare async generators (e.g. in tests) expose
            # `aclose()`.
            if self._stream is not None:
                close = getattr(self._stream, "aclose", None) or getattr(
                    self._stream, "close", None
                )
                if close is not None:
                    await close()

        return False
```

### Issue 2 of 2
posthog/test/ai/test_async_stream_wrapper.py:8-25
`RecordingStream` here, `_RecordingAsyncStream` in `test_openai.py`, and `_RecordingAnthropicStream` in `test_anthropic.py` are structurally identical — each has `_items`, `closed`, `response = "provider-response"`, `__aiter__`/`__anext__`, and `async def close()`. Moving a single shared fixture (e.g. into a `conftest.py` or a small `posthog/test/ai/utils.py`) would remove the duplication and make any future changes a one-line update.

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
posthog/ai/stream.py:55-62
**`aclose()` bypasses provider-stream cleanup**

When a caller does `await response.aclose()` directly (without `async with`), `__getattr__` forwards to `self._generator.aclose()`, which fires the PostHog capture event in the generator's `finally` block — but `self._stream.close()` is never called. The underlying HTTP connection is left open until GC. The `test_aclose_runs_generator_finally_and_captures` test verifies the event fires but intentionally does not assert `source.closed is True`, confirming this gap. Overriding `aclose()` as a proper method (mirroring `__aexit__`'s `try/finally`) would close the provider stream on any teardown path, not only `async with` exit.