perf(xlsx): speed up simple table conversion

[Dex Hunter (dexhunter)]

Fixes #1379.

XLSX conversion keeps pandas’ HTML rendering but handles its simple tables without building a general HTML DOM. This preserves the existing table formatting, with the original HTML converter handling formatting options, embedded images and content the shortcut cannot handle. The escape calls support both current markdownify and its older one-argument API.

On a generated workbook with two sheets of 10,000 rows and eight mixed-type columns each, the final committed-source replay reduced median MarkItDown.convert_stream time from 12.71s to 3.19s (about 75%). Three fresh-process samples ranged from 12.42–12.85s before and 3.13–3.24s after; initialization was outside the timer. This is a representative workload, not the reporter’s workbook. All three timed workloads and ten fidelity/configuration checks matched the original Markdown exactly.

Validation on Linux/Python 3.12:

• pre-commit run --all-files: passed.
• hatch test --python 3.12: 1,001 passed, 14 upstream skips.
• The six new tests also passed against markdownify 0.14.1.

The benchmark used pandas 3.0.6 and markdownify 1.2.3. The Weco run records 11 evaluated candidates plus a separate baseline and two rejected attempts. Candidate 011 (step 13) is the committed choice; its search measurement was 3.21s. The final replay above is separate from that search.

Weco run and candidate history

Developed with AI assistance; the code and tests were reviewed and validated with the checks above.