[timlichtenberg]

Description

Since the per-element closure gate of #39, test_full_chns_converges sits at the edge of what the solver can close (with the Fischer buffer as well as with the new Hirschmann default): for H 1e20, C 1e17, N 1e17 and S 1e16 kg, no restart finds a root that closes S to its own tolerance, and the per-element gate returns the warned fallback after 100 more attempts. Whether a seed converges there depends on round-off, so on the CPU of the runner. The test then ran 2 to 3 times longer, and it timed out at 480 s in the nightly of #43 (https://github.com/FormingWorlds/CALLIOPE/actions/runs/40238902062) while it passed on main (https://github.com/FormingWorlds/CALLIOPE/actions/runs/40241498764). This moves each test away from that edge.

• test_full_chns_converges uses S/H 3e-3, a target that converges, and asserts no fallback, each element within its tolerance, gas plus melt equal to the budget, and the surface pressure.
• A new test keeps a trace-S case far below the edge (S/H 1e-5): it asserts the warned fallback, one warning that names S alone, the attempt count between FALLBACK_ATTEMPTS and nguess, and a state inside the CHNOS mass gate. It makes no assertion on the size of the S miss, which depends on the restart sequence.
• The trace-sulfur fallback test pins S2_bar and S_kg_total at 0 again.
• The example on the installation page used the edge case and promised a few hundred bar; it now uses a converging case and states the output it prints (0.36 bar), and a test runs that code block. The page names CPython 3.12 and 3.13, as pyproject.toml requires.
• The module timeout is 780 s, set from the slowest CI duration by the rule of docs/Explanations/testing.md.
• docs/Explanations/mass_balance.md says that a trace element far below the dominant budget can end in the warned fallback, and what the warning names.

No source change.

Validation of changes

macOS (Apple silicon), Python 3.12, tip 0ffa6e880bf836acb2d09aea5d500ef0d4d7a475:

• pytest -o addopts= tests/test_stoichiometry_integration.py tests/test_solve.py in one process: 78 passed.
• comment-cap --base origin/main --max 3: 0 blocks.
• Nightly on the tip: https://github.com/FormingWorlds/CALLIOPE/actions/runs/40264708201 (all jobs pass).

The targets come from seed scans on macOS and on CI runners of more than one CPU type; the final tests ran only on the nightly runner above.

Checklist

• I have followed the contributing guidelines
• My code follows the style guidelines of this project
• I have performed a self-review of my code
• My changes generate no new warnings or errors
• I have checked that the tests still pass on my computer
• I have updated the docs, as appropriate
• I have added tests for these changes, as appropriate
• I have checked that all dependencies have been updated, as required