fix: vmap-blocker bugs in convert.py and FitEllipse cached_property

Two bugs uncovered when validating fitness._vmap(parameters) on the
new JAX path:

1. autogalaxy/convert.py line ~65: (a == 0) & (b == 0) on JAX scalar
   tracers triggers TracerArrayConversionError via Python's __and__.
   Switch to xp.logical_and so the bool combination stays in JAX-land.
   Fixed all four occurrences: axis_ratio_and_angle_from, two spots in
   shear_magnitude_and_angle_from, and multipole_k_m_and_phi_m_from.

2. autogalaxy/ellipse/fit_ellipse.py: _points_from_major_axis was
   decorated @cached_property. Caching a tracer in self.__dict__
   breaks under vmap (different batch elements share one stale cache).
   Switch to @Property and recompute on access. The numpy path takes a
   ~2x hit on this property but the absolute cost is negligible.

Numpy-path numerics unchanged. 870/870 unit tests pass.

Issue PyAutoGalaxy#411.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: 2 people

autogalaxy/convert.py

@@ -62,7 +62,7 @@ def axis_ratio_and_angle_from(
     """
     angle = 0.5 * xp.arctan2(
         ell_comps[0],
-        xp.where((ell_comps[0] == 0) & (ell_comps[1] == 0), 1.0, ell_comps[1]),
+        xp.where(xp.logical_and(ell_comps[0] == 0, ell_comps[1] == 0), 1.0, ell_comps[1]),
     )
     angle *= 180.0 / xp.pi
 
@@ -202,15 +202,15 @@ def shear_magnitude_and_angle_from(
     """
     angle = (
         0.5
-        * xp.arctan2(gamma_2, xp.where((gamma_1 == 0) & (gamma_2 == 0), 1.0, gamma_1))
+        * xp.arctan2(gamma_2, xp.where(xp.logical_and(gamma_1 == 0, gamma_2 == 0), 1.0, gamma_1))
         * 180.0
         / xp.pi
     )
     magnitude = xp.sqrt(gamma_1**2 + gamma_2**2)
 
     angle = xp.where(angle < 0, angle + 180.0, angle)
     angle = xp.where(
-        (xp.abs(angle - 90.0) > 45.0) & (angle > 90.0), angle - 180.0, angle
+        xp.logical_and(xp.abs(angle - 90.0) > 45.0, angle > 90.0), angle - 180.0, angle
     )
 
     return magnitude, angle
@@ -309,7 +309,7 @@ def multipole_k_m_and_phi_m_from(
         xp.arctan2(
             multipole_comps[0],
             xp.where(
-                (multipole_comps[0] == 0) & (multipole_comps[1] == 0),
+                xp.logical_and(multipole_comps[0] == 0, multipole_comps[1] == 0),
                 1.0,
                 multipole_comps[1],
             ),

autogalaxy/ellipse/fit_ellipse.py

@@ -123,10 +123,10 @@ def points_from_major_axis_from(self) -> "np.ndarray | jax.Array":
         keep = mask_values == 0
         return xp.where(keep[:, None], points, xp.nan)
 
-    @cached_property
+    @property
     def _points_from_major_axis(self) -> np.ndarray:
         """
-        Returns cached (y,x) coordinates on the ellipse that are used to interpolate the data and noise-map values.
+        Returns (y,x) coordinates on the ellipse that are used to interpolate the data and noise-map values.
 
         Returns
         -------