Implement queue-side coalescing (superseding without worker support)

[djmitche]

We extend tasks with:

{
  stack: '<name>',  // requires scope: 'queue:stack:<name>'
  rank: 0, // integer task rank (higher means supersedes), defaults to zero
  ... // existing task fields
  payload: {...},
}

Tasks without task.stack will behave as if they were assigned an exclusive stack name.

claimWork will pick tasks by:

1. Find the highest priority task (FIFO if there are multiple tasks with same priority)
2. If another task with same task.stack and higher task.rank exists, it supersedes the task picked...

In-tree the decision task will use a per-branch string for task.stack if coalescing is supported, and task.rank = push_log_id, to ensure that newer pushes supersedes old pushes.
Example: task.stack = 'gecko:mozilla-central:<label>' where <label> is the task-label.

Note: implementation of query logic like this requires a more advanced database like postgres.

[djmitche]

The current coalescing supports a minimum stack depth and a minimum stack age, both of which serve to prevent tasks from being coalesced when not really necessary (e.g., if two pushes land one after the other, but otherwise we are well within capacity).

[djmitche]

RFC #65 is the postgres migration, a prerequisite to this issue.

[djmitche]

Closing as this hasn't seen any updates in a while. It's always easy to re-open.