fix: SersicCore.intensity_prime ZeroDivisionError on effective_radius=0 (#514)

[Jammy2211]

Summary

SersicCore.intensity_prime raised ZeroDivisionError when effective_radius == 0 because it divided a Python scalar with a plain /. Every other profile (e.g. Sersic.image_2d_via_radii, sersic.py:99/157) divides a numpy array by effective_radius, yielding inf — a non-finite value the non-linear search resamples on — rather than crashing. This switches the scalar division to xp.divide so SersicCore behaves identically (numpy/jax parity; same raise-vs-resample lineage as PyAutoLens#607).

Surfaced by autolens_workspace multi/features/wavelength_dependence/modeling (split out from autolens_workspace#300), whose effective_radius = wavelength*m + c relation — m ~ U(-0.1,0.1), c ~ U(-10,10), both symmetric about 0 — yields effective_radius=0 at the test-mode median instance. alpha was never the cause (Constant 3.0; probe-verified).

API Changes

No signature changes. One behaviour change:

Changed Behaviour

• SersicCore.intensity_prime() (and hence SersicCore / SersicCoreSph image evaluation) with effective_radius <= 0 now returns a non-finite value (inf/nan) instead of raising ZeroDivisionError. Callers that fit SersicCore get the standard non-finite → FitException/resample path rather than a hard crash. The return is now a numpy/jax scalar (was a Python float) for the affected term; numeric results for physical effective_radius > 0 are unchanged (existing exact-value tests still pass).

Migration: none required. Any downstream code that relied on the ZeroDivisionError (there is none in the stack) should instead check np.isfinite.

Test Plan

• test_autogalaxy/profiles/light/standard/test_sersic_core.py — new numpy-only test: effective_radius=0.0 → non-finite intensity_prime(), no raise (3 passed)
• test_autogalaxy/profiles/light/ — 94 passed
• Full test_autogalaxy/ — 1004 passed
• Downstream: autolens_workspace multi/features/wavelength_dependence/modeling runs green under the CI smoke env (workspace PR follows)

Downstream

Workspace follow-up (autolens_workspace, ships after this merges): add the missing auto-simulate block to the multi-wavelength script + drop its no_run.yaml marker.

Generated by the PyAutoLabs agent workflow.

[Jammy2211]

Workspace PR: PyAutoLabs/autolens_workspace#306 (merges after this, library-first).