Run one batch of pending migrated uploads through the pipeline.

Only PROCESSING_SKIPPED rows are dispatched, named or not — dispatch_migrated_file_processing claims the row with a guarded UPDATE — so re-posting the same ids is a no-op rather than a second pipeline run over documents that already had one.

A download-only document is refused by name rather than left to the guard, since "the pipeline is never run over this format" and "this one has already had its batch" would otherwise both read as not_pending, and only the second is worth re-checking.