Refactor ingestor job system

[consolethinks]

The ingestor job system currently can spawn an infinite number of go routines whose only job is to wait until a worker can consume their job on the ingestion task channel. Maybe this needs to be changed to avoid running out of memory.

The bigger issue right now is that if we start using access tokens sent by users for ingestion tasks, then those tasks can't wait for too long. This needs to be addressed in some way (eg. checking if there's a worker available before accepting a task)

[phwissmann]

For the first issue, we can either just use a workpool like Pond https://github.com/alitto/pond or a more elaborate task system/scheduler. I think a workpool is the minimum we should do.