* Fix deadlock and race condition in proposal fetching

* Change from broadcast channel to multi-consumer channel. This means
  only one fetcher task will receive each proposal to be fetched, which
  is the actual behavior we want. Before, with broadcast, we had multiple
  fetchers always fetching the same proposal, which is why we saw race
  conditions causing database serialization errors. It should now be
  possible to reenable multiple workers.
* Use an unbounded channel. This prevents a deadlock where a consumer
  sends back into the channel (e.g. to recursively fetch the parent of
  the proposal it had just fetched), but the channel is full, blocking
  the consumer, the very task responsible for clearing the blockage.

* Add metrics for proposal fetcher