Try an exact ZIP entry match first, then fall back to a case-insensitive lookup for EPUB resources when archive entry casing differs from manifest paths.
Keep ambiguous case-only duplicates exact-match only, and add a regression test that opens a real EPUB through DocumentLoader.