fix(seed): make inherited infra match the CTI stack, not the lensing stack

[Jammy2211]

Phase 1 follow-up on the autocti_assistant epic (PyAutoBrain#136). Makes the
newborn's inherited infrastructure match the CTI stack instead of the
lensing stack it was cloned from.

Why

The lightweight-seed birth copied autolens_assistant's "generic" tier verbatim.
Two of those files carry lensing-stack assumptions that break on CTI — the same
generic-vs-domain boundary leak Phase 0 fixed for file classification, one
layer deeper: a workflow's install step and an audit's module list are domain
content, not framework. This is only surfaceable by an actual birth.

What changed

.github/workflows/wiki-currency.yml — the install step. It ran
pip install autocti from PyPI (the autolens shape, where the library is
released). For CTI that:

• never installs arcticpy — a hard import of autocti but deliberately not a
  pip dependency (source-only sdist; needs libgsl-dev + cython; its own
  requirements downgrade numpy below 2.0), so the install exited 3 and the checks
  never ran;
• pulls a pre-resurrection 2024 wheel, because the CTI release train is not
  yet wired (draft/release/autocti/cti_release_train_wiring.md, human-required)
  — grading the modern docs against it would pass vacuously. Leg 3's own
  contract is "sources clone @ main, doc-pin truth, not the release wheel."

Now clones the main source trees first and installs from them + the arcticpy
recipe (mirrored verbatim from the proven
autocti_workspace_test/.github/scripts/smoke_install.sh), keeping a pinned-PyPI
path for a future release-time workflow_call. PyAutoGalaxy is no longer cloned.

autoassistant/audit_skill_apis.py — the stack module lists. autogalaxy
was in them (autolens sits on autogalaxy; autocti does not — its declared
deps are autoconf/autoarray/autofit). Probing it made --check-version exit
not_installed for a correctly configured CTI user at session start — the check
AGENTS.md runs every code session. Removed from the four hardcoded stack tuples.

Verification

• --check-version now reaches the honest "no baseline — run --write-baseline
  first" instead of "import failed / autogalaxy missing". The remaining redness
  is the missing api_audit_baseline.json, which is a Phase 2-3 content
  artifact (written once real skills are validated against the stack) — not
  infrastructure.
• --scope all still passes 10/10; the module compiles; the workflow is valid YAML.

⚠️ This PR's wiki-currency check is expected to stay RED — honestly

Even with the install fixed, --check-version fails on no baseline (a Phase
2-3 artifact). Writing a baseline now, against a bare seed with no validated
skills, would be gaming the gate — a vacuous green. So the check stays red
until content lands. What this PR does prove, readable in the run log, is that
the install step now succeeds (arcticpy + source stack) and the failure has
moved from broken-infrastructure to content-not-yet-authored.

clone-boundary should pass (skips — the newborn is not a clone reference).

Boundary finding (carried to #136, for a systemic reference fix)

autolens_assistant's "generic" tier has absorbed lensing-stack assumptions
(autogalaxy specifically; the PyPI-install shape; _bootstrap_skill.md and
maintainer.md name PyAutoGalaxy). The durable fix is reference-side — these
files need to be domain-parameterized or reclassified so future births
(autogalaxy_assistant, …) don't each re-hit this. Filed as follow-up on #136.

Part of #136 (Phase 1 completion). Newborn is private; publish waits on Phase 2-3.

[Jammy2211]

CI confirms the fix works. boundary passes (the newborn correctly skips — Phase 0's guard, live in a real cell). wiki-currency's install step now succeeds — the stack imports with no arcticpy/autogalaxy error — and 3 of 5 checks pass (symbol audit, idiom, provenance).

The 2 remaining failures are both content, not infrastructure, and neither is fixable here:

1. --check-version → no baseline (api_audit_baseline.json is a Phase 2-3 artifact; writing one against a bare seed would be a vacuous green).
2. --check-citations → skills/_style.md cites autocti_workspace:scripts/imaging/start_here.py — the substitution fixed the repo name but scripts/imaging/ is the lensing-workspace layout; autocti_workspace uses dataset_1d/imaging_ci. Same boundary leak, content-side.

Both are Phase 2-3 / reference-systemic. This PR's job — moving leg 3 from broken infrastructure to content-not-yet-authored — is done and verified.