Add python-3.12.10-scientific-slim runenv variant

[PaulNewling]

Summary

• New python-3.12.10-scientific-slim runenv variant bundling polars-lts-cpu==1.33.1, numpy==2.2.6, scipy==1.15.3, and pyarrow==21.0.0.
• Registered in pnpm-workspace.yaml and the catalogue package's entrypoints/deps.
• Changeset: patch on the new variant, minor on the catalogue package (new entrypoint added).

Motivation

platforma-open/titeseq-analysis CI currently fails from disk exhaustion on ubuntu-latest while extracting the ~800 MB TensorFlow wheel shipped by the base 3.12.10 runenv (~2.5 GB total across 36 ML packages). That block does pure Hill-equation fitting against MiXCR output — it does not use TF, torch, cudf, transformers, etc. This slim variant ships ~100 MB of wheels that match titeseq's existing software/src/requirements.txt pins exactly, so runtime pip can resolve them under the hardcoded --no-index --find-links={runenv}/packages/ flags in install-deps.tpl.tengo.

Variant layout

python-3.12.10-scientific-slim/
├── package.json  (mirrors python-3.12.10-h5ad/package.json structure; version 1.0.0)
└── config.json   (4 pinned deps, same schema as h5ad/sccoda)

Notes

• No changes to existing variants or the base runenv.
• Future consumers with identical scipy/numpy/polars/pyarrow stacks could migrate off the heavy base (e.g., paratope-clustering, cell-type-annotation, differential-expression — each shares the same runenv and will hit the same wall once they add real integration tests).

[gemini-code-assist]

Code Review

This pull request introduces a lightweight python-3.12.10-scientific-slim run environment variant (~100 MB) bundling polars-lts-cpu, numpy, scipy, and pyarrow. Feedback suggests adding pandas for enhanced utility and aligning the config.json schema with the base environment for consistency.