fix: Sph profile inverse transform name-check asymmetry - #557

Summary

EllProfile.transformed_from_reference_frame_grid_from checked self.__class__.__name__.startswith("Sph") while its mirror transformed_to_reference_frame_grid_from checks endswith("Sph"). Spherical profiles are suffix-named (IsothermalSph, NFWSph, ...) and no concrete class starts with "Sph", so the inverse transform's spherical branch never fired — every spherical profile took the elliptical rotation path. Numerically benign today (spherical profiles have angle = 0, an exact identity rotation), but the transforms were asymmetric and the inverse did wasted rotation work on the hot path. Fixed to endswith("Sph"); a stack-wide sweep (PyAutoGalaxy/PyAutoLens/PyAutoArray/PyAutoFit) confirms this was the only occurrence of the pattern.

Known residual: IsothermalSphMLR (name ends "MLR") is missed by both name checks and continues down the elliptical path — numerically correct, flagged here rather than inventing a new dispatch mechanism for one class.

Closes #555.

API Changes

None — internal changes only.

Test Plan

• New regression test test__sph_named_profile__both_transforms_use_translation_only_path asserts (via mock) that neither transform calls the rotation helper for a suffix-named spherical profile. Control-verified: fails on the unfixed source, passes with the fix.
• test_autogalaxy/profiles: 619 passed.
• Full test_autogalaxy suite run at ship time (counts in PR checks / ship log).

Ship gate note

Shipped with explicit human authorization under Heart RED — reason release validation FAILED (stage integrate), an unrelated in-flight release-validation run; this change was uncommitted at the time of that verdict. Merge remains a separate human decision.

Generated by the PyAutoLabs agent workflow.