CSL fork needs updating toward upstream: scipy ≥1.15 crash + likelihood/sacc predates real-space support #316

#309 builds cosmosis-standard-library into the image, pinned to sachaguer/cosmosis-standard-library@b26fa7f. Two problems, one root cause — the fork is ~373 commits behind upstream:

1. scipy ≥1.15 import crash (all real-space likelihoods). The image's scipy is 1.17, and the fork predates upstream's compatibility fix: scipy.special.lpn was removed, and shear/cl_to_xi_fullsky/legendre.py still imports it — reached by 2pt_like (and sacc_like) via spec_tools at load time, so any real-space inference in the image crashes. Upstream fixed this in a8a941d5. #309 now carries the fix as a build-time patch (docker-patches/csl-scipy-lpn.patch) and an in-image test (test_csl_modules.py) that imports every template-referenced module, so this class of drift fails CI instead of shipping.

2. likelihood/sacc predates real-space support (blocks the native-SACC path, #255). The fork's sacc_like.py default_sections has only harmonic types — it rejects galaxy_shear_xi_plus outright. Upstream main (530e2b1b) understands real-space types and carries the lpn fix. But the UNIONS pipelines also need fork-only modules (sample_S8, the z-dependent linear-alignment interface), so no existing checkout can run the native-SACC ξ± pipeline: it needs modules from both.

The durable fix is updating the fork toward upstream (merge upstream main, re-pin in the Dockerfile); the patch in #309 then retires. Sacha's call on timing/mechanics — happy to prepare the merge as a PR to the fork if useful.

— Claude (Fable) on behalf of Cail.