[scks1]

Summary

Two small, independently reproduced defects on 0.1.9, plus one minor
robustness guard:

1. pdf: _to_markdown_table truncated ragged rows. Column widths came
   from zip(*table) and rows were formatted with zip(row, col_widths);
   zip stops at the shortest row, so cells beyond it were dropped in
   both directions (a long data row lost cells, and a long header lost
   header cells).

2. html/markdownify: a newline in a title attribute was emitted
   literally. convert_a and convert_img escaped only " in the
   title, producing a literal newline inside the quoted title and breaking
   the link/image syntax. convert_img already collapsed newlines in
   alt; the title was missed.

3. pdf (minor): a non-positive page width dropped form tables.
   _extract_form_content_from_words compared line widths against
   page.width * 0.55; with a degenerate width of 0 every row counted as
   a paragraph and the form content was dropped. This is a robustness
   guard — the trigger requires a degenerate MediaBox, and page text
   still survives via the extract_text fallback.

Changes

• _pdf_converter.py: pad rows to the widest row before computing column
  widths in _to_markdown_table; fall back to page_width = 612 when
  page.width is missing, zero, or negative.
• _markdownify.py: collapse \n to a space in the title in
  convert_a and convert_img, as convert_img already did for alt.

Scope note

This PR originally also fixed notebooks whose cell source is a single
string (valid nbformat v4) losing their title in the IPYNB converter.
That fix and its tests have been withdrawn: #2768 covers it with an
identical mechanism, and #2549 covers a broader form of it. The
remaining components — the PDF table pad, the page-width guard, and the
markdownify title fix — have no competing PR.

Tests

Six new tests (seven cases; the page-width test is parametrized over 0
and −5) across tests/test_html.py and tests/test_pdf.py, each
failing on 0.1.9 and passing with this change: both ragged-table
directions, both title-newline cases, and the non-positive page width.

Full run of tests/test_docx.py, tests/test_pdf.py,
tests/test_module_misc.py, and tests/test_html.py on the narrowed
tree: 338 passed, 11 skipped.

Disclosure: this contribution was prepared by Árbol, an AI assistant working under the direction of the account holder (scks1), who set the acceptance criteria and verified the results before submission.