Sandbox Fatal fixes 01.2024, round 2

[dginev]

This PR is a (final) follow-up to the just merged #2309 .

I address another handful of Fatal issues, and one minor binding issue:

• allow deeper recursion in digesting reads of parameters
• allow deeper recursion in stringify
• Guard against infinite loops in complex listings
  • The examples were in cases that deal with QUOTED_RE, but I think there are more general possible pitfalls related to colnum not getting incremented and sanity-checked in every possible arm of the loop conditional.
  • Finding all of the tricky corner cases is a little hard without a deep dive, so I added a distinctive Error message. We can collect all papers exhibiting the issue after a rerun and build a tougher test suite.
• IEEEtran.cls.ltxml needs to support an optional arguments for its list envs, and also rebind the usual env names. This fixes https://github.com/arXiv/html_feedback/issues/417
• \@finalstrut and \@textsuperscript were being used, added to LaTeX.pool.
• A very tricky case in pgfmath was that a single dot . is recognized as zero. So much so that \pgfmathparse{.-.2ex} is a valid expression.

    \documentclass{article}
    \usepackage{pgf}
    \begin{document}
    \pgfmathparse{.-.2ex}\pgfmathresult
    \end{document}

• IEEEtran.cls also does not depend on \sc internally, and a handful of arXiv papers in that class redefined \sc to be something completely different, wreaking some Fatal havoc. I made two IEEEtran-specific primitives to avoid such clashes.
• There was a very unpleasant infinite loop with OmniBus loading natbib late, during \bibitem expansion. For some reason not entirely clear to me the \bibitem DefMacro from natbib didn't stick, and the OmniBus definition expanded into itself in a loop.
  • I created a natbib-specific \lx@nat@bibitem, and used a let to bind \bibitem to it during natbib.sty.ltxml load, which appears to avoid the problem.
  • Last and least, I double-checked $text is always defaulted to an empty string before used in regexes for MathML.pm, which avoids a flurry of pointless perl warnings.

Happy to rerun the sandboxes once we merge here, and triple-check we are in good shape for a release tag.

[dginev]

The changes of the 2 recent stability PRs appear to improve conversion stability in our sandboxes, bringing roughly 1% improvement to the Fatal severity. Details at: https://gist.github.com/dginev/abb72c989f4474a3a33300ed23ac6c86

Ready for review.