Remove ackable_work; ack immediately instead

[mkeeter]

A long time ago, one task accumulated ackable jobs into Downstairs::ackable_work and a different task periodically read out that set and sent acks back to the guest.

The tasks have long since been consolidated, but the ackable_work set remained.

This PR removes the ackable_work set, acking jobs immediately when they are ready. It's a significant LOC simplification, because it's removing intermediate state (which previously had to be checked in unit tests).

There's one subtlety in handling live-repair: previously, the live-repair check would look in ackable_work to see if the next job was ackable, which indicates that live-repair can continue. Now, we update the LiveRepairState directly in Downstairs::ack, storing a result for the pending live-repair job.

[leftwo]

We had a separate task (way way back) in part to make progress if queues were full and we needed to make forward progress. I don't believe that situation still exists.