The testSchedulingTwiceOnTimerCancelsFirstOne test sets an alarm for 1 second in the future and attempts to overwrite it with another timer. The cancellation that is involved with this goes through gRPC, and it involves scheduling work on the gRPC event engine. It's possible that on super slow machines (e.g. with TSAN or other sanitizers enabled), that 1 second isn't enough time for cancellation to catch the timer. We should extend this alarm to be indefinite to avoid this race.