PR #10180 added some tests checking that when the latex subprocess used
by usetex fails, a RuntimeError does propagate up to the Python process.
It originally passed the CI, but since its merge, various CIs have been
failing on the test that a text with $22_2_2$ (an invalid tex
construct) fails when usetex is set.

This is because while latex --interaction=nonstopmode does exit with
error code 1 when processing a file with that construct, the error is
"benign" enough that it still writes a dvi before exiting (the contents
are simply as if the input was $22_{22}$). The dvi goes into the tex
cache and later CI runs pick up the cached dvi instead of running the
subprocess, thus failing the test.

The solution is to use --halt-on-error on top of nonstopmode, see e.g.
https://tex.stackexchange.com/questions/258814/what-is-the-difference-between-interaction-nonstopmode-and-halt-on-error (and to clear your tex cache).

Note that before the PR, there would similarly be the weird behavior
(for the user) that the first attempt to use a "benignly" invalid tex
construct would trigger an exception, but latter attempts would not (for
the same reason as above).

attn @efiring
labeling as rc as it's a CI issue.

PR Summary

PR Checklist

• Has Pytest style unit tests
• Code is PEP 8 compliant
• New features are documented, with examples if plot related
• Documentation is sphinx and numpydoc compliant
• Added an entry to doc/users/next_whats_new/ if major new feature (follow instructions in README.rst there)
• Documented in doc/api/api_changes.rst if API changed in a backward-incompatible way