feat: add numpy ellipse jax_likelihood reference scripts

[Jammy2211]

Summary

Add the numpy-baseline reference scripts under scripts/jax_likelihood_functions/ellipse/ that lock in the reference numbers for ellipse fitting before the JAX path lands. Step 2 of 7 in the ellipse_fitting_jax feature decomposition (PyAutoPrompt/z_features/ellipse_fitting_jax.md).

Each script auto-simulates the dataset on first run, builds a fixed-parameter ellipse model (plus multipole for multipoles.py), and prints every component of every FitEllipse produced by AnalysisEllipse.fit_list_from(...). Prompt 7 of the feature will JIT-wrap analysis.fit_from and assert against these numbers with np.testing.assert_allclose(rtol=1e-4).

Numpy-only — no import jax anywhere in the new scripts. Both fit.py and multipoles.py are deterministic across runs (verified by diff-comparison of two runs).

Reference numbers captured at the prior medians of the model:

Component	fit.py	multipoles.py
log_likelihood	-382.22660947	-725.51150750
chi_squared	764.45321894	1451.02301500
noise_normalization	-27.02295060	-26.84755137
figure_of_merit	-368.71513417	-712.08773182

Issue #41.

Scripts Changed

• scripts/jax_likelihood_functions/ellipse/__init__.py — new empty package marker.
• scripts/jax_likelihood_functions/ellipse/simulator.py — new; simulates a 50×50 single-Sersic-galaxy Imaging dataset (noise_seed=1) into dataset/ellipse/jax_test/. Independent of imaging/simulator.py so the ellipse reference numbers don't drift with future imaging-side retuning.
• scripts/jax_likelihood_functions/ellipse/fit.py — new; single-ellipse numpy reference, prints the four FOM components per fit, ends with a # TODO(7_analysis_ellipse_jax.md) placeholder block.
• scripts/jax_likelihood_functions/ellipse/multipoles.py — new; same shape but with EllipseMultipole(m=4, multipole_comps=(0.05, 0.0)) per ellipse. Uses the multipoles=[[multipole]] list-of-lists composition shape so AnalysisEllipse.fit_list_from's instance.multipoles[i] access works correctly.

Test Plan

• Smoke tests pass for autogalaxy_workspace_test
• Both fit.py and multipoles.py re-runs produce byte-identical reference numbers

🤖 Generated with Claude Code