test: assert every Colab project installs all three samplers

[Jammy2211]

Summary

Adds the regression test that #164 shipped without.

The Colab bootstrap installs its package list with pip install *packages --no-deps, so a sampler that is not named in the list never lands — nothing
arrives transitively via autofit. dynesty and emcee were missing, and
every Colab notebook running af.DynestyStatic or af.Emcee died with
ModuleNotFoundError at fit time, across all six _PROJECTS entries
(observed in HowToFit chapter 1 tutorials 4, 5 and 6).

#164 (merged, 8336939) fixed the list. It did not add a test:
TestRegistry asserted only spec["packages"][0] == "autonerves", so nothing
stopped a future edit dropping a sampler again — which is exactly how the bug
arose. This PR closes that gap.

TestRegistry.test_every_project_installs_every_sampler asserts that for
every key in _PROJECTS ("autofit", "autogalaxy", "autolens", "howtofit",
"howtogalaxy", "howtolens") the resolved packages list contains dynesty,
emcee and nautilus-sampler. It matches on the requirement name with
any version specifier split off, so a future re-pin does not break it.

Release caveat — the main point of this PR's context. Colab notebooks
pip install autonerves fresh on every run, so #164's fix reaches users only
once a new autonerves is released to PyPI. Merging this PR (or #164)
changes nothing user-visible; Colab users keep hitting the
ModuleNotFoundError until that release ships.

autonerves/setup_colab.py is untouched by this PR. The existing
emcee>=3.1.6 / dynesty==2.1.5 pins track PyAutoFit's pyproject.toml and
are left exactly as #164 shipped them; the _SAMPLERS regrouping floated in
the issue plan was deliberately not taken, as unrequested churn on
freshly-merged code — the test asserts on the resolved packages, so it
holds either way if that refactor is ever made.

--no-deps runtime-requirement check

Because of --no-deps, each named sampler's own runtime requirements must
already exist in a stock Colab image:

Package	Runtime requires	In stock Colab?
dynesty==2.1.5	numpy, scipy (declares no install-requires in its metadata)	yes
emcee>=3.1.6	numpy (h5py/scipy only under the extras extra)	yes
nautilus-sampler==1.0.4	numpy>=1.18.0, scipy>=1.4.0, scikit-learn>=0.22.0, threadpoolctl>=3.0.0	yes (threadpoolctl arrives with scikit-learn)

Nothing further needs adding to the install list.

Flagged follow-up (not fixed here)

setup_colab.py pins nautilus-sampler==1.0.4, while PyAutoFit's
pyproject.toml declares nautilus-sampler==1.0.5. Every other sampler
specifier in the list matches autofit exactly, so this one looks like drift
rather than intent. Because the install is --no-deps, Colab therefore gets
1.0.4 against an autofit that expects 1.0.5.

Deliberately left alone: this PR is test-only and the new assertion matches on
package name, so it passes either way. Worth its own prompt.

API Changes

None — test-only changes.

Test Plan

• python3 -m pytest test_autonerves/test_setup_colab.py -q → 21 passed
• python3 -m pytest test_autonerves -q → 184 passed (183 before this PR)
• Control test: dropping each of dynesty, emcee, nautilus-sampler
  from a copy of the registry makes the new assertion fail (it is not a
  tautology); substituting dynesty==9.9.9 and a bare unpinned emcee
  keeps it passing (a future re-pin will not break it).

Closes #165.

Generated by the PyAutoLabs agent workflow.

🤖 Generated with Claude Code

https://claude.ai/code/session_011DMhAYrJ6m2spkTVEWcSKg