Run async code-research benchmarks with a coding agent

Our rewrite with extra checks. The poster didn't test it.

Goal: Benchmark cmarkgfm against popular Python markdown libraries and write a feature and performance report.

Inputs (fill in or confirm):
- Libraries to include (e.g. markdown, mistune, markdown-it-py, commonmark)
- Test documents or sizes to use
- Output folder name (python-markdown-comparison)

Rules:
- Read each library's source and docs to build a feature table (GFM tables, footnotes, extensions, safety).
- Pin versions and record them in the report.
- Run each benchmark several times on the same inputs and report median and spread.
- Generate charts as PNG and include the script that made them.
- Note the machine, Python version and input sizes used so results can be reproduced.
- Check each library's output on a sample document matches cmarkgfm's, and flag differences, so speed isn't compared on wrong output.

Checkpoint: Don't publish or push results; keep everything in the new folder.

If you get stuck: If a library won't install, note it and continue with the rest. If a login, CAPTCHA or two-step code appears, stop and ask me to handle it; never guess or store passwords.

Verify before you finish: Re-run the benchmark once to confirm numbers are stable within a reasonable margin.

Output: A README report with feature table, charts, methodology and how to reproduce.

Done when: someone can rerun your script and get similar numbers.