test(viz): __Visualization Sanity__ pilot block on imaging viz_jit (fast_viz A')

[Jammy2211]

Summary

Adds the first __Visualization Sanity__ regression-net block on autolens_workspace_test/scripts/imaging/modeling_visualization_jit.py. The block exercises the JAX-jittable zero_contour critical-curve path on a deterministic SIE tracer (einstein_radius=1.2, ell_comps=(0.1, 0.0)) at script start, asserting:

• Correctness: non-empty tangential critical curve, finite positive Einstein radius.
• Perf: warm-call latency on the same LensCalc instance is under 100 ms.

Together these catch the three failure modes behind the 2026-04-19 and 2026-05-16 visualization-default reverts: (i) silent-zero source plane from a JAX-trace exception swallowed by the broad except, (ii) Einstein-radius latent collapse, (iii) closure cache-busting in _critical_curve_list_via_zero_contour that made warm calls cost ~10 s instead of ~67 ms. Local validation under PYAUTO_TEST_MODE=2: PASS Visualization Sanity (correctness): 1 tangential CC, einstein_radius=1.1938 and PASS Visualization Sanity (perf): warm call 79.0 ms.

The block runs unconditionally before the Nautilus search, so it fires on every run_all_scripts.sh invocation on the build server.

Scripts Changed

• scripts/imaging/modeling_visualization_jit.py — appends a __Visualization Sanity__ prose block + correctness + perf asserts after the existing Part 1 MGE caching probe. Uses a separately-constructed SIE tracer so the assertions are deterministic regardless of the script's MGE prior medians.

Upstream PRs

• PyAutoGalaxy: perf(lens_calc): cache (f, ZeroSolver) for zero_contour critical curves PyAutoGalaxy#434 (perf cache fix that the warm-call assertion depends on)
• PyAutoLens: fix(viz): make _compute_critical_curve_lines failures loud, not silent PyAutoLens#527 (broad-except tighten — companion safety fix)

Test Plan

• Local smoke run of the script under PYAUTO_TEST_MODE=2 against the worktree libraries — Sanity block passes (einstein_radius=1.1938, warm call 79.0 ms).
• Workspace_test smoke list — 11 / 12 active scripts PASS. The one failure (jax_likelihood_functions/point_source/point.py, JAX vmap returns -1e99 sentinel instead of -83.38) reproduces identically on canonical main without these changes; it is a pre-existing point-source-JAX regression and is out of scope for this PR.
• After PyAutoGalaxy #434 + PyAutoLens #527 merge and a release ships, confirm this script's Sanity block continues to pass against the released library on the next run_all_scripts.sh build.

Closes PyAutoLabs/PyAutoGalaxy#433 (alongside the library companion PRs).

🤖 Generated with Claude Code