fix: support location-based books and handle updated Kindle Cloud Reader UI

[emwbae]

Summary

Two bugs that prevent extraction from working on a significant number of Kindle books:

1. Location-based books produce 0 pages — parsePageNav() in playwright-utils.ts already handles the "Location X of Y" footer format, correctly returning { location, total }. However, the main extraction loop in extract-kindle-book.ts only checks pageNav?.page, which is undefined for location-based books. This triggers the break on the very first iteration, producing an empty result with 0 captured pages.

2. goToPage() crashes on current Kindle Cloud Reader — Amazon renamed the menu item from "Go to Page" to "Go to Location" and changed the underlying element structure. The hardcoded selectors (ion-item[role="listitem"] with hasText: 'Go to Page', input[placeholder="page number"], ion-button[item-i-d="go-to-modal-go-button"]) all fail to match, causing a timeout that kills the extraction before any pages are captured.

Changes

Location-based book support (main extraction loop)

• Added const currentPage = pageNav?.page ?? pageNav?.location to fall back to location when page number is undefined
• Scoped the totalNumContentPages bounds check to only apply when pageNav.page is defined — location-based books don't have a meaningful page-to-content-page mapping, so the original pageNav.page > totalNumContentPages check would never fire anyway (it was the pageNav?.page === undefined check above it that caused the early exit)
• Updated all downstream references (screenshotPath, pageChunk, assertion messages) to use currentPage instead of pageNav.page

Resilient goToPage() navigation

• Broadened menu button selector with .or() fallbacks for both ion-button and plain button elements
• Changed menu item matching from exact 'Go to Page' string to regex /go to (location|page)/i to handle both old and new Kindle Cloud Reader UI
• Added fallback selectors for the page/location input field (placeholder*="page" and placeholder*="location")
• Added fallback selectors for the Go button (text match + legacy item-i-d attribute)
• Wrapped the entire function in try/catch — on failure, dismisses any open modals via Escape keypresses and logs a warning instead of crashing the entire extraction
• Wrapped the end-of-extraction goToPage(initialPageNav.page) reset in try/catch to prevent a crash after all pages have already been successfully captured

Reproduction

1. Open any Kindle book that displays "Location X of Y" in the footer (rather than "Page X of Y") — this is common for many self-published and KDP titles
2. Run npx tsx src/extract-kindle-book.ts
3. Extraction completes instantly with 0 pages captured, because parsePageNav() returns { location: X, total: Y } (no .page property), and the loop immediately breaks on pageNav?.page === undefined
4. After this fix, the same book correctly captures all pages via location-based navigation

Related issues

• Partially addresses Error running extract-kindle-book.ts #14 — same class of selector breakage from Kindle Cloud Reader UI updates
• Partially addresses Issue: Stuck on Last Page During Screenshot Capture #6 — the try/catch on the end-of-extraction goToPage() call prevents the "target page closed" crash that occurs when the browser context is closing

Known limitations

The downstream export scripts (export-book-markdown.ts, export-book-pdf.ts, export-book-audio.ts) use tocItem.page and content.findIndex((c) => c.page >= nextTocItem.page!) to map chapters to content chunks. For location-based books, the PageChunk.page field now contains location numbers rather than page numbers, which may cause chapter boundary mismatches in exports. This is a pre-existing limitation (these scripts couldn't work at all before since extraction produced 0 pages) and would be best addressed in a follow-up PR.

Validated

• Tested on a location-based book — 181 page screenshots captured successfully
• goToPage() gracefully recovers when selectors don't match (Escape key dismisses modals)
• Full pipeline works end-to-end: extraction → transcription (GPT-4.1-mini) → markdown export
• Passes prettier and eslint (pre-commit hooks ran successfully)

[cif49]

Fyi, as of writing this this branch still works, the main one does not.
Thank you.