there is a race condition between setting up the timer and waiting for the failpoint. if the timer elapses before getting to the continuation where we wait for the failpoint, we'll just end up waiting forever.

we could increase the timeout to make this less likely, but this will (a) slow down the test, and (b) not necessarily prevent this from happening in the future. perhaps an existing (or new) failpoint can be utilized, or some other means of testing the correct abort behavior without explicitly relying on a timeout.