Adopt the deferred private EPUB ingestion prototype into the canonical Reader #289

Recovered product debt

Current README and the historical #274 delivery record explicitly retain unpublished EPUB/batch Workbench work. It was not part of the accepted UTF-8 Reader integration (#273/#276), nor of #288's report comparator. Open/closed issue and PR searches found no delivered EPUB integration. The prototype bytes were not recovered in this session; the historical claim is not implementation or test evidence.

User outcome

Analyze a locally supplied, unencrypted EPUB with the same canonical metrics, without uploading the book, executing embedded content or silently treating a container file as its literary body.

Acceptance

• First locate and inspect any surviving handoff; deduplicate against current source. Recover only useful code into the canonical ingestion layer, not a parallel Workbench/analyzer.
• Explicit supported EPUB subset and reading-order policy; separate raw archive, extracted body and normalized-text identities and versions. Preserve editions/translations as distinct inputs.
• Bounded archive/member counts, compressed and expanded bytes; reject traversal, duplicate/ambiguous members, encrypted input, unsupported structures and unsafe XML. Never execute scripts, follow remote resources, write archive paths or bypass DRM.
• Reuse the actual analyzer and current privacy/error/create-only output semantics; short-text representativeness remains explicit.
• Real installed CLI -> HTML/JSON/CSV tests with small original synthetic fixtures: reading order, Unicode, missing spine, malformed archives and size/path attacks. No full copyrighted books in fixtures or reports.
• Document the accepted input subset and failures. No corpus admission, FantLab-input match, new backend or parity claim. M2 remains 0/5.

Independent of #280 browser report comparison. Keep execution ordering in the existing compact project queue; do not create a competing worker or speculative integration branch.