docs: the outputs and checking guides run on a sample project

[jmineau]

The guides that open a project now run when the docs are built, on a small sample project. Part of #116.

The sample project is four receptors an hour apart on the morning of 15 January 2021, run on the same cached HRRR files as the Getting Started page. It is built the first time a page asks for it, about 40 s once per docs build, and copied into a temporary folder for each page that uses it, so a page's files never land in the repository. just docs-serve keeps it between rebuilds. docs/_ext/docs_examples.py holds the setup, and the Getting Started page uses it too.

What runs now:

• The outputs guide: the footprint, particle and simulation maps, the facet and availability plots, the footprint's receptor, grid and settings, the NetCDF round trip, the particle table, and selecting many simulations, each with the output it prints. Its examples use the sample's day in place of July 2023.
• The checking guide: project.status() and the particle and footprint maps of "My footprint looks wrong". The sample has no failures, so the failed table is empty, and the page says so.
• The receptors guide: the point, column and multipoint examples end with receptor.plot.map(). These need no simulation.

Running them corrected three things. rid, in the outputs guide, was never defined, so the line that used it could not run. The mean footprint on a linear scale is one bright cell on a dark square, because most cells are 0; it is now drawn on a log scale, as the trajectory-analyses guide does. And receptor.plot.map() labeled the guide's own column example, whose heights are above sea level, with "m AGL", and a multipoint receptor's colorbar always said "Height AGL". Both labels now use the receptor's reference (own commit, two tests that fail on the old code).

Left as plain code: the particle file path, whose settings folder name is an example, the scene selection, which needs a column the sample does not have, and the empty-footprint example, since the sample has none. The trajectory-analyses guide needs your own measurements and a second met, so it is untouched.

Cost. A clean just build-docs takes about 165 s on a compute node, against about 95 s with the Getting Started page and 60 s before. just quality-check (1,285 tests, one fewer per example that now runs instead of being compiled) and all the pre-commit hooks pass.

Three layout problems in the plots these pages now show are in #219: the legend of sim.plot.map(), the label of project.plot.availability(), and the facet title. I left them, since the layout is a design choice.

Based on #218: its two commits come first, and the rest are this PR's. I will rebase it onto main when #218 merges.

[jmineau]

Rebased onto main now that #218 is merged: the five commits of this PR, no conflicts, the same changes as before (+265 -51 in 9 files).

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

Files with missing lines	Coverage Δ
src/stilt/visualization.py	94.79% <100.00%> (ø)