config: default critical_curves_method to zero_contour

[Jammy2211]

Summary

Flip autogalaxy/config/visualize/general.yaml default critical_curves_method from marching_squares to zero_contour.

Motivation

For slope=2 isothermal lens profiles, marching_squares fabricates a noisy polygonal radial caustic where the radial eigenvalue never cleanly crosses zero (analytically it stays at 1; numerically ellipticity makes it dip near zero, which marching squares picks up as spurious contours). The resulting jagged octagon is not at the physically meaningful 2→1 image transition — it's a sampling artifact.

zero_contour correctly omits the degenerate radial caustic for isothermal profiles and renders clean smooth ovals when one physically exists (verified with slope=1.8).

Runtime: ~2–3× slower per visualize call for the small integration-test dataset, but this is plot-only code — not hit per likelihood evaluation.

API Changes

None — internal config default only. Users who explicitly override critical_curves_method in their own visualize/general.yaml are unaffected. Default behavior changes: caustics/critical curves are now computed via the JAX zero-contour method instead of marching squares.
See full details below.

Test Plan

• pytest test_autogalaxy/ passes (834/834)
• autolens_workspace_test/scripts/imaging/visualization.py runs end-to-end; all assertions pass; radial caustic cleanly omitted for slope=2, smooth oval shown for slope=1.8.

Full API Changes (for automation & release notes)

Removed

None.

Added

None.

Changed Behaviour

• Default value of visualize/general.yaml::general.critical_curves_method changed from "marching_squares" to "zero_contour".
• Consequence for downstream plotters (_caustics_from, _critical_curves_from in autogalaxy/util/plot_utils.py and autogalaxy/plot/plot_utils.py): they now call tangential_caustic_list_via_zero_contour_from() / radial_caustic_list_via_zero_contour_from() and the corresponding *_critical_curve_list_via_zero_contour_from() methods by default, instead of the marching-squares grid-based variants.
• For singular power-law profiles (e.g. isothermal, slope=2), the radial caustic is correctly omitted. Previously marching squares would draw a spurious polygonal artifact.
• For profiles where a radial caustic physically exists, it is rendered as a smooth continuous curve rather than a pixel-aligned polygon.

Migration

• No user action required for the common case.
• To restore the previous behavior, set critical_curves_method: marching_squares in a project-local config/visualize/general.yaml.

🤖 Generated with Claude Code

[Jammy2211]

Workspace PR: PyAutoLabs/autolens_workspace_test#29