Add an error-reporting endpoint and associated queue

[hannahwhy]

If a fetch script fails on an item, there is currently no way to report failure back to the tracker. As a result, the item remains in the "claimed" list.

As failure may be due to either a fetch script error or some property of the item (e.g. strings with unpermitted characters being used as subdomain names), we need some way to examine these items in isolation.

Adding an endpoint to report an error (POST /errored, maybe) as well as a new error queue will help with these goals:

1. The impact on the existing design is minimal. We're adding queues, not changing existing ones.
2. Fetch scripts do not have to use the error-reporting endpoint, so all existing scripts will continue to work as they are.
3. This prevents "dangerous" items from being issued to clients.

[hannahwhy]

A clarification on the error-reporting endpoint's behavior:

If the posted item exists in the claims list, it is removed from the claims list and placed in the error queue. (I guess we could check nickname, too, but I'm not sure that's really worth it -- it's trivial to defeat that check.)

If the posted item does not exist in the claims list, the request is either ignored (e.g. 200 returned but nothing happens) or a 4xx error is issued (probably 404).