refactor: reduce magic in transform decorator

[Jammy2211]

Summary

Reduce the implicit "magic" in the @aa.grid_dec.transform decorator while keeping its DRY benefits. This is Phase 1+2 of the refactor described in #271.

Phase 1: The is_transformed flag now lives as a property on the grid object itself (grid.is_transformed) instead of being threaded through **kwargs. This makes transform state explicit, inspectable, and impossible to lose through missed kwarg propagation.

Phase 2: Added rotate_back parameter to @aa.grid_dec.transform(rotate_back=True). When enabled, the decorator automatically rotates vector outputs (e.g. deflection angles) back from the profile's reference frame to the observer frame — eliminating the need for deflection methods to manually call self.rotated_grid_from_reference_frame_from(...).

API Changes

• @aa.grid_dec.transform now accepts an optional rotate_back=True keyword argument. Existing usage without arguments is fully backward-compatible.
• grid.is_transformed is now a public readable/writable property on all grid objects (Grid2D, Grid2DIrregular, Grid1D).

See full details below.

Test Plan

• 725 PyAutoArray tests pass
• 834 PyAutoGalaxy tests pass (including 12 new regression tests)
• Exact numerical regression values locked in for Isothermal, NFW, PowerLaw

Full API Changes (for automation & release notes)

Added

• AbstractNDArray.is_transformed — public property (getter/setter) exposing transform state on grid objects
• transform(rotate_back=True) — optional parameter on the transform decorator for automatic back-rotation of vector outputs

Changed Behaviour

• @aa.grid_dec.transform now reads/writes grid.is_transformed instead of kwargs["is_transformed"]. The kwarg-based approach is no longer used.

Migration

• Before: kwargs.get("is_transformed") to check transform state
• After: grid.is_transformed to check transform state
• No action needed for most users — the decorator handles this internally

🤖 Generated with Claude Code