refactor: drop sip_tpv, rely on astropy for TPV WCS (closes #713)

[cailmdaley]

Summary

Removes the sip_tpv dependency from shapepipe entirely. Modern astropy (≥5) parses TPV distortion natively via WCSLIB, as do SExtractor, PSFEx, and every other downstream WCS consumer in the pipeline, so the keyword-level pv_to_sip conversion that split_exp used to run was no longer buying us anything.

Dropping the conversion lets us drop sip_tpv itself — an unmaintained 2017 package whose v1.1 imports pkg_resources at module load and is therefore incompatible with setuptools>=81. That's the actual root cause of #713 (rewritten to reflect this). PR #714's setuptools<81 pin is a symptom workaround that becomes unnecessary once sip_tpv is gone.

What changed

Code

• split_exp_package/split_exp.py: drop import sip_tpv as stp and the stp.pv_to_sip(h) call; remove the now-dead transf_coord parameter from process/create_hdus; fix the module docstring.
• split_exp_runner.py, python_example_runner.py: remove sip_tpv from depends=[…].
• find_exposures_runner.py: remove sip_tpv from depends=[…] — spurious declaration, find_exposures_package never imported it.

Environment

• environment.yml, environment-dev.yml, Dockerfile, install_shapepipe: drop sip_tpv.
• docs/source/dependencies.md: drop the sip_tpv row.
• docs/source/refs.bib: drop the orphan shupe:12 entry.

Tests (new)

• src/shapepipe/tests/test_split_exp.py: three unittests that build a synthetic multi-CCD FITS with TPV distortion, run SplitExposures, and verify:
  1. Per-CCD output headers' WCS matches the source exposure's WCS at several pixel samples.
  2. The headers*.npy side-output round-trips pixel → world → pixel.
  3. flag output files are stored as int16.

All four modules in src/shapepipe/tests/ still pass locally (7/7).

Not touched

• scripts/jupyter/wcs.ipynb still imports sip_tpv — it's an exploratory notebook last edited ~2.5 years ago. Left alone to keep the diff scoped.

Interaction with open PRs

• fix: small develop bugs (#709, #711, #712, #713) #714 (develop-bugs) currently pins setuptools<81 as a workaround. Once this PR merges, that pin can come out. If fix: small develop bugs (#709, #711, #712, #713) #714 merges first the pin becomes dead code; happy to follow up with a cleanup commit in either direction.
• test: Tier 0-2 scaffolding (shell syntax, imports, entrypoints, runner metadata) #708 (testing scaffold) introduces tests/unit/ and includes an xfailed import test for shapepipe.modules.split_exp_package.split_exp. That test should xpass once either this PR or the setuptools<81 pin lands.

Reviewer Checklist

• The PR targets the develop branch
• The PR is assigned to the developer
• The PR has appropriate labels
• The PR is included in appropriate projects and/or milestones
• The PR includes a clear description of the proposed changes
• If the PR addresses an open issue the description includes "closes #"
• The code and documentation style match the current standards
• Documentation has been added/updated consistently with the code
• All CI tests are passing
• API docs have been built and checked at least once (if relevant)
• All changed files have been checked and comments provided to the developer

[cailmdaley]

Follow-up: empirical verification of the TPV-downstream premise.

The previous state of this PR relied on an inference — "SExtractor and PSFEx link against WCSLIB, which handles TPV, therefore they handle TPV" — without actually running the external tools. Adding integration tests to actually verify it:

test_tpv_external_tools.py (commit f35cec51)

• SExtractorTPVTestCase — builds a synthetic 512×512 image with a CFIS-scale TPV header (non-trivial radial terms PV{1,2}_{4,6}), injects 9 Gaussian point sources at known pixel positions, runs source-extractor 2.25.0, and asserts each detection's ALPHA_J2000 / DELTA_J2000 agrees with astropy's TPV pix→world mapping.

  • Tolerance: 10 mas (~0.05 px at 0.187"/pix).
  • Observed max separation: 0.5 mas — essentially pure centroiding noise; the WCS math is exact.

• PSFExTPVTestCase — same kind of image → SExtractor FITS_LDAC → PSFEx. Asserts:

  1. LDAC_IMHEAD preserves every PV keyword verbatim through SExtractor (it does).
  2. psfex 3.21.1 exits successfully on a TPV-headed LDAC (it does; 9 stars loaded, 9 accepted, CHI2 ≈ 0.81).
  3. The LDAC-embedded WCS round-trips pix↔world identically to the source header (1e-9 px agreement).

Both tests are gated on shutil.which("source-extractor") / which("psfex") and skip gracefully otherwise.

Combined with the earlier code audit showing every shapepipe module consumes WCS via astropy.wcs.WCS(header) with zero keyword-level introspection, this closes the "what about downstream?" question for me. Ready for review.

[cailmdaley]

Piggybacked a small follow-up: removed the TreeCorr row from docs/source/dependencies.md and its orphan jarvis:04 BibTeX entry from docs/source/refs.bib. Both were missed in #715's cleanup. Same file diff as the sip_tpv/shupe:12 removal; natural home rather than a new PR for 2 changes.

[martinkilbinger]

I ran the two tests and they passed. Not sure how strict these really are, but I guess we can merge this PR. This package seems to be before support in astropy was implemented.