This was broken in matrix-org/synapse#10272, where we made incoming federation event processing asynchronous.

The failure mode was that when we called await_sync_timeline_contains it would get event R and then separately S, where previously we'd get them down in the same batch, and so paginating from the prev_batch token returned R rather than just Q. This didn't happen previously as the call to send_event previously took long enough that by the time the test called await_sync_timeline_contains the server would have process both R and S and so they'd come down in the same batch.

After this has landed we should remove the test from the blacklist.