Run Complement tests sequentially (#11910)

Since #11811 there has been general Complement flakiness around networking.
It seems like tests are hitting the wrong containers. In an effort to diagnose
the cause of this, as well as reduce its impact on this project, set the
parallelsim to 1 (no parallelism) when running tests.

If this fixes the flakiness then this indicates the cause and I can diagnose
this further. If this doesn't fix the flakiness then that implies some kind
of test pollution which also helps to diagnose this further.