readme_toc.py misparses tilde and longer Markdown fences

Summary

The README TOC generator only recognizes backtick fences and treats every line beginning with three backticks as a close. It therefore includes headings from tilde fences and from content inside a longer backtick fence.

Reproduction

Run the README TOC generator on:

## Before
~~~markdown
## Not a heading

After

Also try a four-backtick fence containing a literal triple-backtick line.

Expected: headings inside either fenced block are excluded, and the longer fence remains open until a closing fence with the same marker and at least the same length.

Actual: headings inside tilde fences are treated as headings, and a triple-backtick content line can prematurely close a longer backtick fence.

## Root cause

The parser used a boolean toggled by `line.strip().startswith("```")`; it did not track fence marker type or length.

## PR-ready reference

I prepared a tested reference branch containing a minimal parser fix and regression tests:

- Commit: https://github.com/Hughhhhcoder/codex/commit/57516831cf353aab3a73c4e99136ab987a00d1db
- Branch: `codex/codex-readme-fence`

The patch tracks the fence character and opening length, ignores backtick openers whose info string contains a backtick, and accepts only valid matching closers.

## Validation

`python -m pytest -c /tmp/codex-empty.ini --noconftest -q scripts/test_readme_toc.py` — 3 passed.

Because the current contribution policy does not accept external code contributions or pull requests, I am submitting this analysis and reference commit here instead of opening a PR. Please feel free to implement or adapt the fix.

Start with scripts/test_readme_toc.py and run the documented pytest command, then follow it to the README TOC generator. Confirm the parser handles tilde fences and longer backtick fences as described, with headings inside excluded and the regression tests passing.