fix(inversion): make AbstractMeshGeometry picklable (xp module → _use_jax bool + property)

[Jammy2211]

Summary

Phase 4 subprocess visualization (PyAutoFit #1279) requires sending FitImaging instances over an IPC boundary. A spike against the current FitImaging showed it's unpicklable: AbstractMeshGeometry.__init__ stores self._xp = xp (the literal numpy or jax.numpy module) and Python can't pickle modules. This was the only pickle barrier — the rest of the fit graph (mappers, linear operators, computed matrices, tracer, dataset, mask) serialises fine.

Fix: replace the module-attribute pattern with the _use_jax: bool + _xp property pattern already used in Analysis._xp (PyAutoFit) and AbstractMaker._xp (PyAutoArray decorators, per CLAUDE.md). Eliminates the pickle barrier at the API level instead of working around it with __getstate__/__setstate__ hooks.

API Changes

• AbstractMeshGeometry.__init__(xp=np) continues to accept the same xp= kwarg — no caller changes needed.
• The instance no longer holds _xp as a module attribute; it holds _use_jax: bool and exposes _xp as a @property that returns numpy or jax.numpy on demand.
• All existing self._xp reads continue to work transparently (property has the same interface as attribute access).

See full details below.

Test Plan

• New test_autoarray/inversion/pixelization/mesh_geometry/test_picklability.py — 5 new tests covering numpy + JAX backends across MeshGeometryRectangular and MeshGeometryDelaunay, plus a __init__-state invariant test.
• Existing test_autoarray/inversion/pixelization/mesh_geometry/ tests still pass.
• Wider test_autoarray/inversion/ suite passes (171/171).
• End-to-end: a populated FitImaging (Sersic lens + Rectangular-adaptive-density pixelization source) round-trips through pickle.dumps/loads with log_likelihood Δ=0.00e+00 on both numpy and JAX backends. Pickle size 4637.7 KB.

Full API Changes (for automation & release notes)

Changed Behaviour

• autoarray.inversion.mesh.mesh_geometry.abstract.AbstractMeshGeometry:
  • Instance attribute _xp removed; replaced by _use_jax: bool (set from the xp= kwarg as xp is not np).
  • _xp is now a @property that returns numpy if _use_jax is False else jax.numpy. JAX is imported lazily on access.

Migration

• No required migration. The xp= kwarg on __init__ is unchanged. Any reader of instance._xp continues to work (property returns the same module the old attribute held).
• Any writer of instance._xp = ... (none currently exist in the codebase) would now fail at runtime. Confirmed via grep: zero writes outside __init__.

Why this matters

• FitImaging now pickles cleanly, unblocking PyAutoFit #1279 (Phase 4 subprocess visualization). The single-class fix on the abstract parent covers both MeshGeometryRectangular and MeshGeometryDelaunay.

🤖 Generated with Claude Code