fix: publish as pyautoreduce — autoreduce on PyPI is not ours

[Jammy2211]

Closes #71.

The finding

#71 was filed as "published autoreduce 0.9 never got the Python 3.12 floor". That premise is disproved. autoreduce on PyPI is not a PyAutoLabs package — it is github.com/ayush9pandey/AutoReduce, "Python tools for automated model reduction of nonlinear dynamical systems", with a release history running 0.1.0 (2020-10-15) through 0.9 (2026-08-12).

Corroborating evidence that we have never published it:

• git ls-remote --tags PyAutoLabs/PyAutoReduce returns nothing, and the build is setuptools-scm + dynamic = ["version"] — an untagged tree cannot produce a 0.9.
• PyAutoHands has zero references to autoreduce: not in release.yml's matrix, not in pre_build.sh's WORKSPACE_SPECS.
• This repo has no release workflow of its own.

So there is no floor gap, and the <=3.14.7 cap in #71 is a third party's. requires-python = ">=3.12" from d7bd916a was already correct.

The actual defect

We claimed a distribution name owned by someone else, and told users to install it:

pip install autoreduce            # ← a stranger's package
pip install "autoreduce[hst]"     # ← [hst] doesn't exist there; pip warns, installs anyway

Second-order: autoreduce/__init__.py resolved __version__ via importlib.metadata.version("autoreduce"), and package/provenance.py writes that into every reduction.json. On a machine with the foreign package installed, our provenance recorded its version as ours.

The change

Distribution becomes pyautoreduce (free on PyPI, matches the repo name and the org prefix). The import package stays autoreduce — no source renaming. target.py, psf/starred_epsf.py, scripts/reduce_cosmos_web_ring.py and docs/design/hst_acs_pipeline.md:253 were already writing pyautoreduce[starred], so this resolves existing drift rather than creating it.

File	Change
pyproject.toml	name, with the reason recorded in place so nobody tidies it back
autoreduce/__init__.py	_version("pyautoreduce") — load-bearing, see below
README.md	the three install lines
docs/design/hst_acs_pipeline.md:479	autoreduce[frames] → pyautoreduce[frames]
autoreduce/psf/stpsf_model.py	ImportError install hint
autoreduce/package/cosmic_rays.py	ImportError install hint

Deliberately unchanged:

• autoreduce/package/provenance.py — its tuple is fed to __import__(package).__version__. Those are import names, and the import package is unchanged; renaming that entry would break provenance.
• .github/workflows/main.yml's package: autoreduce — Heart's lib-tests.yml uses it as the import package (pytest --cov ${{ inputs.package }}) and as a repo-mapping key, not as a distribution name.

Verification

Built and installed on a real 3.12 venv:

Name:            pyautoreduce
Requires-Python: >=3.12
top_level.txt:   autoreduce          # import package preserved

autoreduce.__version__  -> 0.0.post1
provenance record       -> {'python': '3.12.3', 'autoreduce': '0.0.post1', ...}
302 passed, 206 warnings in 12.59s

The __init__.py edit is load-bearing, not cosmetic — in the same venv, version("autoreduce") raises PackageNotFoundError, so leaving it would have silently degraded __version__ to the "0.0.dev0" fallback.

0.0.post1 is setuptools-scm's untagged fallback; the real version arrives with the tag, which is the follow-up's job.

Follow-up

Induction into the PyAutoHands release machinery and the family's date-based versioning is filed separately as PyAutoMind/draft/release/pyautoreduce/pyautoreduce_release_induction.md, blocked until this merges. Worth noting there: date versioning needs no packaging change — this repo's [tool.setuptools_scm] block is already byte-identical to PyAutoLens's, and the family's 2026.8.23.1 versions come from the annotated tag release.yml pushes before python -m build.

Companion docs PR in autoreduce_workspace follows behind this one (library-first merge gate).

One thing noticed but not fixed here, to keep the PR minimal: the built wheel's top_level.txt lists docs and scripts alongside autoreduce. That is pre-existing ([tool.setuptools.packages.find] excludes docs but not scripts, and neither exclusion is taking effect) and unrelated to this diff — worth its own hygiene task.

---

Generated by Claude Code