chore: re-pin the API baseline to the released 2026.8.4.1 stack

[Jammy2211]

Fixes the red wiki-currency check now sitting on main.

What broke

wiki/core/api_audit_baseline.json pinned the stack at 2026.7.29.2, generated 2026-08-01. The 2026.8.4.1 release on 2026-08-04 moved all four pinned packages (autonerves, autoarray, autofit, autogalaxy), so Version drift (--check-version) started failing for every PR to this repo.

It surfaced on #10, but was not caused by it — #10 touched two .md files and merged with the red check, inheriting it onto main.

Why this had to be regenerated off-workspace

--write-baseline derives api_surface from the installed library. This workspace installs the PyAuto* source checkouts (autogalaxy 2026.7.23.1), not the released stack CI grades against — regenerating here would have pinned the wrong version and turned the check green on a false premise.

So this was produced in a clean venv with autogalaxy==2026.8.4.1, PYTHONPATH cleared, and every module verified to resolve to venv site-packages rather than a source checkout:

autonerves   2026.8.4.1     site-packages (good)
autoarray    2026.8.4.1     site-packages (good)
autofit      2026.8.4.1     site-packages (good)
autogalaxy   2026.8.4.1     site-packages (good)

One real API change, not just version strings

autogalaxy.plot goes 27 → 28 symbols. The addition is subplot_fit_imaging_list, exported in autogalaxy/plot/__init__.py between 2026.7.29.2 and 2026.8.4.1. Every other namespace hash is byte-identical.

That symbol is not yet documented in the wiki or skills. Surfacing newly exported public APIs is one of this check's stated jobs, so it is worth a follow-up decision on whether it warrants a row or a page — deliberately left out of this PR to keep the diff to the re-pin.

Verification

Control-tested in an environment that reproduces CI, so the result is attributable to the baseline and not the venv:

Baseline	--check-version
old (2026.7.29.2)	exit 1 — reproduces the CI failure
new (2026.8.4.1)	exit 0

All five checks the workflow runs now pass:

• --check-version → 0
• --scope all → 0
• --lint-idioms → 0
• --check-provenance → 0
• --check-citations → 0 (scanned 78 files, 1060 citations, 0 missing)

🤖 Generated with Claude Code

[Jammy2211]

Added: documentation for subplot_fit_imaging_list

The re-pin surfaced autogalaxy.plot going 27 → 28 symbols, and the new export was documented nowhere. Rather than leave that as a follow-up it is now in this PR (78d3163).

It turned out not to be a simple missing row — three existing claims became wrong once the function existed, each checked against inspect.signature rather than read off the release diff:

1. AGENTS.md and wiki/core/api/plotting.md both enumerate which functions accept output_filename. subplot_fit_imaging_list takes it (default fit_combined), so both lists were incomplete. The documented list now matches dir(aplt) exactly, verified by introspection:

   documented: [plot_array, plot_grid, subplot_fit_imaging_list, subplot_imaging_dataset,
                subplot_imaging_dataset_list, subplot_interferometer_dataset,
                subplot_interferometer_dirty_images]
   actual    : (identical)
   MATCH     : True
   

2. skills/ag_plot_fit.md stated flatly that "the fit subplots do not take output_filename". Now false for exactly one function — qualified rather than deleted, and the skill gains a worked example of comparing several fits in one figure.

3. wiki/core/api/plotting.md claimed every subplot takes colormap and use_log10. The _list variants do not. This also caught a pre-existing error: subplot_imaging_dataset_list has had that leaner signature all along and the page never said so.

plotting.md is provenance-stamped, so it is re-pinned to PyAutoGalaxy 13d3023c and re-stamped via --write-provenance. All five wiki-currency checks still pass against the released 2026.8.4.1 stack.