fix(mobile): extract covers from large epub imports

[codedogQBY]

Analysis

• 大文件（40M以上）epub文件导入出现的问题 #412 reports that 40 MB+ EPUB imports on mobile do not show covers, cannot vectorize, and the cover replacement action cannot select images.
• [Bug] 导入后无法显示书籍封面图片 #187 reports the same Android large-file cover extraction failure for 50 MB+ EPUB imports.
• PR fix(mobile): prevent large book vectorization and sync OOM #423 already addresses the large-file vectorization/OOM side and references 大文件（40M以上）epub文件导入出现的问题 #412, but it does not close 大文件（40M以上）epub文件导入出现的问题 #412 because the import metadata/cover path was still skipped for large EPUBs.
• The mobile import path intentionally skipped EPUB metadata extraction above 32 MB to avoid reading the entire file into JS memory. That prevented cover extraction even though the EPUB ZIP directory allows reading only container.xml, OPF, and the cover entry.
• The mobile document picker MIME map also lacked image extensions, so the cover replacement button asked the picker for application/octet-stream instead of image MIME types.

Changes

• Add a range-based EPUB metadata extractor that reads the ZIP EOCD/central directory from the file tail, then loads only metadata and cover entries on demand.
• Keep the existing whole-file metadata extraction path for small EPUBs, but route large mobile EPUB imports through the range extractor instead of skipping metadata.
• Preserve entry size caps for OPF/container text and cover image reads so large books do not get loaded wholesale.
• Add image MIME mappings for jpg, jpeg, png, webp, and gif so mobile cover replacement opens image files correctly.

Verification

• pnpm exec biome check packages/app-expo/src/lib/book/metadata-extractor.ts packages/app-expo/src/stores/library-store.ts packages/app-expo/src/lib/platform/expo-platform-service.ts
  • Passes with existing warnings in touched files (noExplicitAny in Expo platform service and unused saveCoverToAppData).
• pnpm --filter @readany/app-expo exec tsc --noEmit
• git diff --check
• Bundled metadata-extractor.ts with esbuild and ran an ad-hoc Node verification that builds a 40 MB+ EPUB in memory; extractEpubMetadataFromFile correctly returned title, author, and cover bytes without reading through the import code path.

Fixes #412
Fixes #187