docs: describe the MGE sigma floor as the pixel scale, not 0.01"

[Jammy2211]

Phase 2 of the sweep started in autolens_workspace#467 (merged). Paired with autogalaxy_workspace#203.

Why

Both pages told users the MGE sigma ladder spans 0.01" to the mask radius. That was never what the code did — mge_model_from hardcoded np.linspace(-4, ...), a 2-dex discrepancy that PyAutoGalaxy#549 fixed by making the floor configurable. The workspace examples now floor the ladder at a tenth of the pixel scale, so the prose and the code samples were describing something that exists nowhere.

Changed

• skills/ag_basis_profiles.md — ladder prose and code sample updated; the point-model paragraph now names sigma_min and its 0.01" default rather than stating a bare literal.
• wiki/core/concepts/linear_light_profiles_and_mge.md — same, plus a short note on why the lower end is anchored to the pixel scale (it stops the basis spending Gaussians on scales the data cannot resolve).

Not changed

skills/ag_multi_dataset.md matched a 1e-4 grep, but that is an intensity LogUniformPrior bound, not a sigma floor. Left alone.

Provenance

The wiki page is provenance-stamped, so this PR also:

• adds autogalaxy/analysis/model_util.py to its cited PyAutoGalaxy paths — that is where sigma_min lives, and the page now describes it
• re-pins PyAutoGalaxy at 13d3023c (verified to contain #549) and autogalaxy_workspace at 1f821ba (this sweep's phase-2 commit)
• bumps last_updated and re-stamps content_sha256 via --write-provenance

autoassistant/audit_skill_apis.py exits 0 with 0 missing/broken symbols across 46 files and 220 unique symbols.

Note the wiki-currency workflow's paired-PR support applies here: autogalaxy_workspace has a branch of this same name, so the cited-source checkout grades against the paired change rather than against main.

Merge gate

pending-release — sigma_min exists only on PyAutoGalaxy main and in no released PyAutoGalaxy.

🤖 Generated with Claude Code

[Jammy2211]

wiki-currency failure is pre-existing release drift, not this PR

The only failing sub-check is Version drift (--check-version):

- autogalaxy version: 2026.7.29.2 -> 2026.8.4.1 (informational)
The skills/wiki were validated against the baseline. Run `--scope all` to audit
drift and `--write-baseline` to re-pin once fixed.

Evidence it is not caused by this PR:

• wiki/core/api_audit_baseline.json was generated 2026-08-01 pinning autogalaxy 2026.7.29.2. PyPI latest is now 2026.8.4.1, released today.
• This PR touches exactly two files — skills/ag_basis_profiles.md and wiki/core/concepts/linear_light_profiles_and_mge.md — and does not include the baseline.
• Every other sub-check passes, including the two this PR could plausibly have broken:
  • Symbol audit (--scope all): scanned 46 files; 220 unique symbols; missing/broken: 0
  • Provenance (--check-provenance): warnings only (unresolvable git checkouts under a packaged install) — no content_sha256 error, so the re-stamp in this PR is good
  • Idiom deny-list: clean — scanned 122 files, no defunct idioms

Any PR opened against this repo today will hit the same failure until the baseline is re-pinned to 2026.8.4.1.

I have deliberately not re-pinned it here. --write-baseline derives the API surface from the installed library, and my local environment runs the source checkout (2026.7.23.1), not the released 2026.8.4.1 this CI grades against — re-pinning locally would stamp a baseline for the wrong version and turn the check green on a false premise. The re-pin should be done against the released stack, as its own change.