hydrant-2tg was filed when one db.repos row without repo_metadata failed a whole POST /repos/resync batch. 45396ea already fixed _resync to give such a row fresh tracked metadata and a pending entry, but its test only resyncs the orphan on its own. this adds the bead's acceptance case: two tracked repos sitting in resync after a failed backfill plus a relay style row without metadata, resynced in one batch. all three get queued, the orphan gets tracked metadata whose index_id names its pending entry, its row stays as it was, and repos, pending, resync and error_transport counts end at 3, 3, 0 and 0. break-check: with the old metadata lookup (bail on a missing row) the test fails with "repo metadata not found for did:web:listed.example" and none of the batch is queued.