Records where the objects stored before location-truth are (V204).

Every writer records a phoenix_kit_file_locations row for each bucket it stores an object in, and marks its instances checked. Instances stored earlier by writers that did not (Tessera tiles, comment attachments) are unchecked. This job walks them in @batch_size batches, in uuid order, checks each enabled bucket for the instance's key once, records every bucket that has it (Storage.Locations.record/2), and marks the instance checked with how many held it (Locations.mark_checked/2) — a miss too, so it is not work again and the Health count reaches zero.

"Unchecked", not "has no location row": a read that finds a key in a bucket records that one bucket, which must not retire the instance from this walk while its other copies are unrecorded.

Throttled. One batch per run, and the next run is scheduled @pause_seconds later, on the file_processing queue, so a large or remote bucket is walked at a steady pace instead of all at once.

Queued by itself (the maintainer's decision, plan §"Next: V204"): PhoenixKit.Supervisor calls maybe_enqueue/0 shortly after boot, and the daily trash prune does too, whenever any instance has no row. Only one pending run exists at a time (unique over [:worker, :queue], ignoring the cursor). Reads keep working meanwhile: the manager checks the other buckets for a key with no rows, and records what it finds.

run_pass/1 runs a whole pass in the calling process.