fix: re-mirror likelihood scripts and datasets from clean origin/main of _developer

[Jammy2211]

Summary

Phase 1 (#1, merged via PR #2) copied the source-of-truth scripts and datasets from the local canonical checkout of autolens_workspace_developer, which had ~36 modified files and a non-empty dirty staging area at the time. As a result the merged mirror carried in-progress modifications instead of the published origin/main state.

Surfaced by the autolens_profiling F1 triage (autolens_workspace_developer#67): when smoking the same scripts against clean origin/main of _developer, imaging/mge.py PASSED its regression assertion (27379.388907), but the mirrored copy was reporting 27542.08 — a 0.6% drift that turned out to be from copying the canonical's locally-modified dataset/imaging/hst/*.fits, not an upstream drift.

This PR re-mirrors scripts + datasets from a fresh feature/jit-regression-drift worktree based on clean origin/main and re-applies the same Phase 1 path rewrites verbatim.

What changed

• 9 likelihood scripts refreshed (−470 / +294 net): reverts in-progress _developer modifications that should not have been part of the Phase 1 mirror.
• dataset/imaging/hst/{data,noise_map}.fits restored to clean-main bytes (canonical's locally-modified versions were displacing the published values).
• dataset/imaging/hst/tracer.json restored to clean-main JSON.
• dataset/interferometer/hannah/ added (5 files, 828K) — was tracked on clean main of _developer but missed by Phase 1's cp from the canonical (which lacked it locally for whatever reason).

Path rewrites reapplied unchanged:

• Path(\"jax_profiling\") / \"dataset\" → Path(\"dataset\")
• _script_dir.parents[2] → _script_dir.parents[1]
• \"jax_profiling\" / \"results\" / \"jit\" → \"results\" / \"likelihood\"
• Docstring sibling-script and dataset refs rewritten to the new layout.
• if al.util.dataset.should_simulate(...) subprocess block → clean FileNotFoundError pointing at autolens_workspace_developer/jax_profiling/dataset_setup/.

Verification

• Re-smoked likelihood/imaging/mge.py against the refreshed mirror:

  Eager regression assertion PASSED: log_likelihood matches 27379.388907
  Regression assertion PASSED: log_likelihood matches 27379.388907
  

  The 0.6% drift reported in PR feat: Phase 1 — mirror JIT likelihood profiling scripts #2's test plan is fully resolved — it was always a Phase 1 hygiene issue, not a real upstream drift.

• The remaining point_source script drifts noted in PR feat: Phase 1 — mirror JIT likelihood profiling scripts #2 are real upstream behaviour changes, not mirror artifacts. They are tracked separately:

  • PyAutoLabs/autolens_workspace_developer#67 — F1 triage parent.
  • PyAutoLabs/PyAutoLens#514 — upstream PointSolver / FitPositions drift.

  These remain failing in this PR (load-bearing regression assertions) until #514 is resolved.

• Two additional bugs uncovered by smoking all 10 jit scripts on clean main (independent of the mirror hygiene fixed here):

  • PyAutoLabs/autolens_workspace_developer#68 — jit/imaging/pixelization.py JIT vs eager mapping matrix shape mismatch.
  • PyAutoLabs/autolens_workspace_developer#69 — jit/imaging/delaunay.py log_evidence rebuild returns -inf vs FitImaging's finite value.

Test plan

• python likelihood/imaging/mge.py PASSES regression (27379.388907 matches constant).
• (Expected to remain failing) python likelihood/point_source/image_plane.py — load-bearing while PyAutoLens#514 is open.
• (Expected to remain failing) python likelihood/point_source/source_plane.py — load-bearing while PyAutoLens#514 is open.
• (Expected to remain failing) python likelihood/imaging/pixelization.py — load-bearing while autolens_workspace_developer#68 is open.
• (Expected to remain failing) python likelihood/imaging/delaunay.py — load-bearing while autolens_workspace_developer#69 is open.

Refs PyAutoLabs/autolens_workspace_developer#67

🤖 Generated with Claude Code