refactor writer to a writer/worker architecture

Currently, the metadata writer performs two functions:
1. Reads block data from the Stacks index
2. Performs all metadata fetches from queued jobs

To optimize performance and resource use, we should split these so the writer only takes care of (1) while new auto-scalable workers do (2). We could use a Redis queue to enqueue jobs instead of a postgres table like we do today, and it would allow us to down/up scale workers when required to save on infra costs.