fix: synthetic positions fallback in test mode and PYAUTO_SMALL_DATASETS

[Jammy2211]

Summary

Fixes a ValueError: zero-size array to reduction operation fmax in Result.positions_likelihood_from when PYAUTO_TEST_MODE=2 produces unphysical mass models that back-trace to empty / NaN image-plane positions. Also collapses the os.environ.pop("PYAUTO_SMALL_DATASETS") workaround that simulator scripts use around PointSolver.solve by short-circuiting the solver itself when small-datasets mode is on.

Closes #477.

API Changes

No public API signature changes. Two test-mode-only behavior additions, both gated on env vars and inert in production:

• Result.positions_likelihood_from now substitutes [(1.0, 0.0), (-1.0, 0.0)] when PYAUTO_TEST_MODE is set and the resolved positions are empty / NaN / inf.
• PointSolver.solve now short-circuits to [(1.0, 0.0), (0.0, 1.0)] when PYAUTO_SMALL_DATASETS=1 is set, skipping the triangle-tiling solve.

Outside test mode / smoke-test mode neither branch is reached. See full details below.

Test Plan

• New unit test test__positions_likelihood_from__test_mode_fallback exercises the empty-positions path with PYAUTO_TEST_MODE=2.
• Existing test_autolens/analysis/test_result.py and test_autolens/point/triangles/ suites still pass (21 tests).
• Manual smoke: PYAUTO_TEST_MODE=2 python scripts/imaging/features/pixelization/delaunay.py no longer crashes at the original result.py:333 error site (a separate downstream border_relocator.py:450 bug surfaces, tracked separately).
• PointSolver short-circuit is exercised by every simulator integration script run with PYAUTO_SMALL_DATASETS=1 — workspace cleanup PR will remove the os.environ.pop workarounds.

Full API Changes (for automation & release notes)

Removed

• None

Added

• None (no new public symbols)

Renamed

• None

Changed Signature

• None

Changed Behaviour

• autolens.analysis.result.Result.positions_likelihood_from — when PYAUTO_TEST_MODE is set (any non-zero value) and the resolved positions array is empty (shape[0] < 2), or contains NaN, or contains inf, substitutes the synthetic pair [(1.0, 0.0), (-1.0, 0.0)] and emits a single logger.warning before computing the threshold. Outside test mode no change.
• autolens.point.solver.point_solver.PointSolver.solve — when PYAUTO_SMALL_DATASETS=1 is set, returns aa.Grid2DIrregular(values=[(1.0, 0.0), (0.0, 1.0)]) immediately, skipping super().solve_triangles(...) and downstream filtering. Outside small-datasets mode no change.

Migration

• Workspace simulator scripts that currently wrap PointSolver.solve with os.environ.pop("PYAUTO_SMALL_DATASETS", None) / restore can now delete that workaround. A follow-up workspace PR removes it from scripts/group/simulator.py, scripts/group/features/multi_gaussian_expansion/simulator.py, and scripts/group/features/no_lens_light/simulator.py.

🤖 Generated with Claude Code

[Jammy2211]

Workspace PR: PyAutoLabs/autolens_workspace#102