This run_test helper sets up a convention that allows
us to give really short tracebacks for errors, and
eventually we can have more control over running
individual tests. (The latter goal has some
complications, since we often intentionally leak
setup in tests.)
This reverts commit bcdd12773e.
We need to do some improvements in handling FetchStatus for initial
narrows before this will be safe to deploy in production.
If individual messages arrive before we get the latest
messages from the server, they can create gaps in rendering,
and would often be offscreen anyway, so we just ignore them.