fix: preserve reading order for styled inline text in doc.html() (#4000)

[Praddy2006]

Description

Fixes #4000 — text extracted from PDFs generated via doc.html() could
appear out of reading order when the source HTML mixed plain text with
styled inline elements (<strong>, <em>, <span>, etc.). This happened
because html2canvas paints DOM nodes in CSS painting order, which does not
always match DOM/reading order, and jsPDF wrote each text draw to the PDF
content stream as it was painted.

Changes

• src/modules/context2d.js: When autoPaging === 'text', text draws are
  now buffered (textRunBuffer) instead of written immediately. A new
  flushTextRuns() method sorts buffered entries into top-to-bottom,
  left-to-right reading order (grouping by line via a y-tolerance) before
  writing them to the content stream.
• src/modules/html.js: flushTextRuns() is called automatically after
  html2canvas rendering completes in toContext2d_post().
• The existing 'slice'/true autoPaging mode is untouched — buffering
  only applies to 'text' mode, since 'slice' mode isn't affected by this
  issue and direct Context2D API callers may not always route through
  doc.html()'s flush hook.

Testing

• Added test/specs/html-text-order.spec.js with two cases:
  • a single sentence containing <strong>/<em> inline runs
  • four paragraphs exercising different mixed-styling patterns (plain,
    <span>-split, nested inline, whole-phrase bold), asserting they
    extract in correct top-to-bottom order
• npm run test-local — 467 specs, 0 failures

Checklist

• Test cases added
• npm run prettier run before commit
• No dist/ files included in this PR
• No unrelated changes