Handle non-fatal task failures by re-adding task to queue

... Also set a TTL to the task.

The issue body does not name files, tests, or an entry point. Start by tracing how non-fatal task failures are handled and how tasks are re-added to the queue; identify the existing queue mechanism before determining how task TTL is configured. Done means failed tasks are re-queued with an expiry and the behavior is covered by a relevant test.

Written by the indexing model from the issue text.