Improve "batch-died" autoclean

Every death callback fires S::Batch::DeadSet#clean. This is not a problem if you have a few dead batches but becomes cumbersome if the user has lots of dead batches for some reason. The more you use batches, the worse the performance gets.

1. Put an hourly throttle on the operation so we don't try to clean more often than once per hour.
2. Sidekiq Pro 9.0 should increase the Redis requirement to >=7.4 which allows us to use expiring hash fields. We should migrate the current Set to a Hash where each BID member is a Hash field with an expiry matching the batch's expiry.