This happened because we cache Task instances, not the values, and a failed task is still a *valid* task that can repeatedly be awaited. This lead to the Task being saved even if it failed, and the same exception constantly being re-thrown.
This fix invalidates the cache if it finds a failed Task, and lets it retry fetching as normal (hopefully successfully this time).
This also refactors a large portion of the DI toolchain, since
I discovered that you shouldn't be reusing IDbConnection objects.
Instead, most services and stores are now declared transient, and
the webhook cache has been moved to a database-independent storage
singleton by itself.
