Clean up the strike stream over time #127
Description
The stream of strikes currently just grows without bound. This means that while a client or worker is starting up, they'll have more work to do and they will need to process all prior strikes.
Worse, the clients could be in an inconsistent state briefly while they're catching up on old strikes: for example, if you strike task_a then later restore task_a, there's a brief moment in time when a client is catching up on strikes where they may see that task_a should be stricken.
There's also no way to just clear the strike list completely.