ci(docs): stop installing an unused TeX stack; bound the docs job

[thodson-usgs]

Summary

The docs build on main ran 150 minutes instead of its usual ~6. It was not a code regression and not the notebooks.

The commit it ran on (#356) touched only CI config and pyproject.toml — git diff f7c248b7 deba1331 -- docs/ demos/ dataretrieval/ is empty, so the build inputs were byte-identical to the run before it, which took 6m28s.

Timing the two runs against each other localizes it precisely:

phase	fast run (6m25s)	hung run (150m)
apt download (127 pkgs)	0.3 min	143.8 min
notebook execution	3.7 min	3.7 min

Notebook execution was identical. The entire difference was apt fetching the same 314 MB at an effective 37 KB/s off a slow azure.archive.ubuntu.com mirror. Single packages stalled for over 20 minutes each (mupdf-tools 23.6 min, openjdk-21-jre-headless 22.2 min).

What this changes

1. Drop the TeX stack (~266 MB of the 314 MB). An HTML build never reaches it — no math extension is configured in docs/source/conf.py (neither mathjax nor imgmath), and nothing renders math to images. latexmk, texlive-*, and dvipng were downloaded and never executed, yet accounted for 5 of the 6 slowest single-package waits. This removes the tail risk rather than trimming it: with pandoc alone there is no longer a large download to be slow.

pandoc is kept — nbsphinx shells out to it for notebook markdown cells, and the build fails with PandocMissing without it (verified).

2. Add timeout-minutes: 30. The next slow mirror will find whatever is left, and the default is 6 hours. This matters more than it appears: runs are serialized on the ref (concurrency, cancel-in-progress: false), so an unbounded run doesn't just lose its own docs — it holds the queue for every commit behind it. That is why this surfaced as "docs are stuck on main" rather than as one slow run. A cell-level nbsphinx_timeout was considered and deliberately left out, since notebook execution was never the problem here.

Test plan

• sphinx-build -b html with TeX binaries off PATH → build succeeded
• All 44 non-notebook HTML pages byte-identical to a build with TeX present (examples/ pages differ only because notebooks re-execute against live data)
• Zero math*.png in either build — confirms nothing was rendering math via LaTeX
• Confirmed pandoc is genuinely required (removing it fails with PandocMissing)
• Workflow YAML parses; executed commands contain no texlive/latexmk/dvipng
• Confirm this PR's own docs run lands in the normal ~5 min range

🤖 Generated with Claude Code