Add deadletter queue for causal diffs

[ChrisPenner]

Overview

A recently submitted contribution included references to unknown builtins, this caused the contribution diff to crash, but the queue didn't have any error handling, so it would keep retrying without backoff and ended up gumming up the diff queue and also using up all our Postgres credits :'(

This adds error handling which should catch invalid diffs and mark them as errors so they're not retried.

• Run the migration on staging and prod

Implementation notes

• Catch exceptions in diff transactions
• Mark these entries with an error
• Don't pull erroring diffs from the queue

Test coverage

I'll test one known invalid diff on production to ensure it's all working correctly.

Loose ends

I'd prefer we had a proper queueing solution; ideally one with auto-backoff, a GUI interface, and maybe even some form of alerting. Maybe we can look at adding one.