feat: cluster plots honor the configured critical-curve engine #807

Affected Repositories

• PyAutoLens (only)

Branch Survey

No active.md claim on PyAutoLens.

Suggested branch: feature/cluster-curves-engine-dispatch

Implementation Steps

1. autolens/cluster/plot/cluster_plots.py:
   • Add a small helper that maps (method, kind ∈ {tangential, radial} × {critical curve, caustic}) to the LensCalc routine. For marching_squares that is tangential_critical_curve_list_from and the existing routines; for zero_contour it is tangential_critical_curve_list_via_zero_contour_from and its siblings (autogalaxy/operate/lens_calc.py ~1500–1700).
   • Read the method once per plot call via autogalaxy.util.plot_utils._critical_curves_method().
   • Mirror how autogalaxy/util/plot_utils.py (~470–540) and autolens/lens/plot/critical_curves.py call the zero-contour variants (their arguments: grid, initial points, etc.).
   • Keep _lens_calc_for_plane as is.
2. Docstrings: describe the engine choice. Keep the marching-squares grid note.
3. Tests in test_autolens/cluster/plot/ (or beside the existing cluster plot tests), on a two-source-plane cluster fixture:
   • monkeypatch the selector to each engine and spy on the LensCalc methods called, for tangential and radial curves and caustics, per plane;
   • assert the plane-1 and plane-2 geometries differ;
   • set the config to zero_contour while jax_zero_contour import fails, and assert the marching-squares fallback (the selector warns).
4. Run the focused PyAutoLens tests and the full suite. Workspace cluster visualisation smoke, if available.

Key Files

• autolens/cluster/plot/cluster_plots.py (_lens_calc_for_plane ~167, plot_critical_curves ~346, plot_caustics ~440)
• PyAutoGalaxy autogalaxy/util/plot_utils.py:405 (_critical_curves_method) and ~470–540 (how other plots dispatch)
• PyAutoGalaxy autogalaxy/operate/lens_calc.py ~1431–1720 (*_via_zero_contour_from)
• autolens/lens/plot/critical_curves.py

Out of scope

Context-aware JIT dispatch, docstring fixes elsewhere, cluster-scale accuracy and performance, and the jax_zero_contour triage: these are the remaining phase-3 scope.

PyAutoMind draft/feature/autolens/cluster_curves_engine_dispatch.md (Epic: cluster-strong-lensing, Phase: 3a, Consequence: judge).

Make autolens/cluster/plot/cluster_plots.py select the existing configured critical-curve engine for each source plane in plot_critical_curves and plot_caustics. Preserve LensCalc.from_tracer(use_multi_plane=True, plane_j=j) and per-plane colours, labels, radial option and source-plane mapping. The default remains marching_squares; explicit zero_contour uses the existing LensCalc zero-contour methods. Use PyAutoGalaxy's existing private method selector so config and missing-dependency fallback agree with other plots. Keep changes in PyAutoLens only.

Verify both configured engines on a two-source-plane cluster fixture, including tangential and radial curves and caustics. Tests must assert method routing and distinct per-plane geometry, not merely saved image paths. Check the configured-zero-contour behavior when the dependency is unavailable. Run focused PyAutoLens tests and appropriate workspace cluster visualization smoke. Do not claim JIT or gradient safety from a plotting test.