fix(rag): vectorize chapters from epub toc structure

[codedogQBY]

Summary

• Add a reusable RAG helper that maps EPUB TOC leaves to logical chapter section ranges, using all TOC nodes as boundaries.
• Update desktop book extraction to vectorize logical TOC chapters instead of treating every spine section as a chapter.
• Return a human-readable number from ragToc while keeping index as the value to pass into chapter tools.

Analysis

Before this change, desktop vectorization iterated book.sections and created one ChapterData per spine section. That means front matter, volume title pages, split XHTML files, and other non-chapter sections could become AI-visible “chapters”. For EPUBs with multi-volume TOCs, asking about a numbered chapter could lead the model to inspect the wrong section or a fallback Section N title instead of the actual book structure.

The new grouping follows the EPUB TOC: leaf TOC entries become logical chapters, parent entries such as volume nodes are used as boundaries, and multiple spine sections between chapter anchors are merged into one vectorized chapter. If an EPUB has no usable TOC anchors, extraction falls back to the previous one-section-per-entry behavior.

Fixes #411

Verification

• git diff --check
• pnpm --filter @readany/core test -- chapter-structure tools
• pnpm --filter @readany/core exec tsc --noEmit
• pnpm --filter app exec tsc --noEmit
• pnpm exec biome check packages/core/src/rag/chapter-structure.ts packages/core/src/rag/chapter-structure.test.ts packages/core/src/rag/index.ts packages/core/src/ai/tools/rag-tools.ts packages/core/src/ai/__tests__/tools.test.ts packages/app/src/lib/rag/book-extractor.ts (passes with existing noExplicitAny warnings in legacy test/extractor code)