Retire in-code catalogue versioning and the pre-Snakemake run layers

[cailmdaley]

Retire in-code catalogue versioning and the pre-Snakemake run layers

Branch: split/retire-versioning → develop. 4 commits, 58 files, +607 / −8112. Every deleted file is on develop at 240b37e4.

Why

ShapePipe writes shear catalogues; it never reads, reprocesses, or stays compatible with a v1.x run, so the code has no business knowing which catalogue version it produces — git tags are the version. It still carried the v1.x era: the v1.3/v1.4 → 7, v1.5 → 8, v1.6 → 9 patch tables, -V/-P flags, P1..P9 loops, the post-hoc summary/ + missing_job_* scrape, and two generations of CANFAR bash whose job the Snakemake workflow does. After this PR no code path branches on a catalogue version.

pipeline/ and modules/ held no versioning — only three dual-layout has_option switches (EXP_BASE_DIR, ME_DOT_PSF_EXP_DIR, ME_IMAGE_EXP_DIR) whose legacy branches were reachable from example/cfis/defunct/ alone. The mass is run infrastructure.

Retired

What	Replacement
ccd_psf_handler.py, get_ccds_run.py, get_ccds_with_psf.py, build_and_plot_coverage_maps.sh — the version table and its VERSIONS loop	the exposure-count map (#887)
utilities/summary.py, summary_params_pre_v2.py, summary_run, summary_tiles.py, summary_run.ipynb, post_processing.md — per-patch scrape, n_patch = 8 hardcoded	manifests/, logs/, completeness.py, sp report
collate_star_cat.py: -V/-P patch loop, and the -p mccd branch (focal-plane CCD ids, Jacobian rotation of external-mccd moments)	the PSFEx path, verbatim; the workflow refuses psf_model: mccd
CANFAR gen 0/1: canfar_submit.py, canfar_submit_job, init_run_exclusive_canfar.sh, job_sp_canfar.bash, combine_runs.bash, distribute_tiles.py, link_to_exp_for_tile.py, merge_final_cat.py, … ; skaha	SLURM executor, prepare/exposure/tile.smk, canfar_monitor
merge_headers/psfex_interp/vignetmaker runners' log_run_sp.txt branch	the sharded store only; a config without EXP_BASE_DIR/ME_*_EXP_DIR fails loudly

The header-download / corner-extraction coverage chain is not retired here: #887 replaces it with the exposure-count map and removes it there.

Deliberately kept

• Gen-2 bash (run_job_sp_canfar_v2.0.bash, job_sp_canfar_v2.0.bash, functions.sh → command()): sp_validation's image_sims.smk calls it. Version-free; retires with the image-sims port.
• create_final_cat.py: writes the patches/<run> group sp_validation reads; renaming lands in lockstep there.
• completeness.py, ngmix_range.py, tile_local(): their content is part of rule params, so edits rerun finished units; comment edits wait for a campaign boundary.

Questions

1. Is anyone still launching production on CANFAR? The workflow has no CANFAR mode.
2. check_tile_coverage.py claims an sp_validation caller we can't find — live?
3. merge_final_cat.py was the only merger for per-tile randoms; needed for the tomographic round?
4. Does anyone still collate MCCD validation catalogues with collate_star_cat.py -p mccd?

Merging with #887: trivial conflicts in pyproject.toml [project.scripts] (drop all four entry points), tests/module/test_coverage.py (take #887's) and the workflow/README.md scripts list (keep both).

Tests: tests/unit tests/workflow tests/module/{test_coverage,test_psfex_interp,test_vignetmaker,test_pipeline,test_run}.py — 489 passed, 1 skipped in the dev image.

— Claude (Opus) on behalf of Cail

[cailmdaley]

From the smk-g7 acceptance run (integration of #879 + #886 + #887 + this branch, 64 tiles / 127 exposures): coverage_map ran and completed — 4891 CCDs with a PSF, 54.5 deg² of depth-map footprint, and the depth sampled at each catalogue object reproduces the catalogue's own NGMIX_N_EPOCH histogram (mode 3; 99.987% of objects at depth > 0). That is the acceptance run this PR named as not yet done. One tuning note: the job requested 31.25 GB and used 1.09 GB (3%); worth sizing from the footprint count like the other campaign-level rules do.

— Claude (Fable) on behalf of Cail