Import complete audiobook payloads and review per-book file mappings

[bandoracer]

Completed downloads previously supported only one book file. This change imports complete audiobook chapter sets and sidecars from exact client inventories, and sends uncertain or multi-book payloads to explicit per-file book mapping with a required destination preview.

• Preserve selected/unselected/unknown inventory state; reject traversal, symlinks, duplicate resolved paths, incomplete files and conflicting metadata. SABnzbd uses successful completed history output, never archive filenames as extracted-file proof.
• Extend durable manifests with per-file book identity (append-only migration 0032). Commit all chapter records/links together, resume published files after failure, preserve chapter order on retries, and verify the whole set before cleanup.
• Add a responsive mapping review, stale-preview rejection, persistent Skip/Reject and Reopen. Explicitly retained required files block source deletion.
• Keep existing single-book Calibre handoff; manual-path/Calibre crash recovery, replacement and temporary-stage reclamation remain follow-up work.

Validation: full Go race suite with disposable Postgres, Go vet, six frontend tests, production build, 17 desktop/mobile browser checks (one mobile-only case inapplicable on desktop), and diff checks pass. Locally built ARM64 containers passed missing-payload review, forced database failure, process-restart recovery, complete multi-disc/cover import, rescans, isolated database restore, and the forms/Basic/none auth matrix. The packaged test uses a read-only qBittorrent HTTP fixture; real-client/live homelab qualification remains pending.

Stacked on #3. This PR is unreleased, does not publish PR images, and does not certify the full S01–S25 plan or live deployment.