Unique tasks should be unique per queue

[thomasst]

When removing a unique task from a queue, we check other states (e.g. scheduled / error / etc.) before removing the task object from Redis to make sure the task is not referenced anywhere else. However, we don't check queues with other names, we just check the current queue. I propose to include the queue name in the task ID hash to prevent issues with removing unique tasks. Alternatively, we'd have to check all other queues for tasks before removing a unique task, or have some kind of reference counter.

This is potentially a breaking change.