Fix text loss at page boundaries during incremental parsing

[alfarhan]

Summary

All format parsers (TXT, Markdown, EPUB, FB2) silently drop text when a paragraph spans a page-batch boundary during incremental parsing (hot extend). This manifests as missing sentences/paragraphs when reading — the user turns the page and text that should follow is gone.

Root Cause

When layoutAndExtractLines() is called and the page batch limit (maxPages) is reached mid-paragraph:

1. The processLine callback stops adding lines to pages (returns or sets a flag)
2. But the extract loop continues — extractLine() destructively erases words from the ParsedText object
3. After flushBlock returns false, the ParsedText is destroyed/reset — remaining words are permanently lost
4. The next parsePages call resumes from a saved file offset that is past the lost text

This affects every parser because they all share the same pattern of calling layoutAndExtractLines() without an abort callback to stop word consumption when the page limit is hit.

The Fix

Part 1 — Stop consuming words when the page limit is hit:

Pass an abort callback to layoutAndExtractLines() that returns true when the page batch limit is reached. The existing abort check at ParsedText.cpp:286-287 already handles this correctly — it returns false without consuming more words. The problem was that no caller was passing this callback.

Part 2 — Preserve unconsumed words for the next batch:

When the page limit interrupts a paragraph mid-layout, the ParsedText object still contains the remaining words. Instead of destroying it, save it as a class member (pendingBlock_ / pendingTextBlock_) and flush it at the start of the next parsePages call.

Parsers Fixed

Parser	Changes
PlainTextParser	Added pendingBlock_ member, abort callback in flushBlock, resume logic
MarkdownParser	Added pendingTextBlock_ member, abort callback in flushTextBlock and memory-flush path
Fb2Parser	Added abort callback, conditional currentTextBlock_ reset
ChapterHtmlSlimParser	Changed includeLastLine from false to true in emergency split (abort callback was already passed)

Test plan

• Built and flashed on Xteink X4 (ESP32-C3)
• Tested with long Arabic RTL markdown files that previously exhibited missing text between pages
• Verified text flows correctly across all page boundaries after clearing page cache
• Test with EPUB and FB2 Arabic content
• Test with LTR content to verify no regression
• Test under memory pressure (large files on device with limited free heap)

🤖 Generated with Claude Code