feat: oversampled PSF support in SimulatorImaging + workspace example

[Jammy2211]

Overview

Follow-up to the oversampled-PSF series (design PyAutoArray#353; chosen as option (a) at the phase-3 sign-off, autolens_workspace#232): support simulation with an oversampled PSF and land the user-facing simulator.py example with it. Run mode: supervised --auto.

Plan

• PyAutoGalaxy — OperateImage.convolved_padded_image_2d_from(grid, psf, xp) (shared by Galaxies and Tracer): build the padded all-false mask (as padded_grid_from, using psf.kernel_shape_image_resolution), a padded grid with uniform over-sample size s (the existing padded_grid_from pads the border with 1s, which the uniformity guard correctly rejects), evaluate unbinned on its over-sampled coordinates, convolve at the fine resolution (blurring_image=None — all flux in-frame, same as today's padded flow) and return the image-resolution convolved padded image. via_galaxies_from switches on psf.convolve_over_sample_size > 1.
• PyAutoArray — SimulatorImaging.via_image_from(..., image_is_convolved: bool = False): skip the internal convolution when handed an already-convolved image. Additive, default preserves behaviour.
• PyAutoLens — the same switch in via_tracer_from (it duplicates the PyAutoGalaxy pattern); trims with kernel_shape_image_resolution.
• autolens_workspace — extend scripts/imaging/simulator.py with the oversampled-PSF example (fine kernel, convolve_over_sample_size, uniform over-sampling requirement).
• Tests — unit: s=2 simulation equals brute-force fine construction; s=1 regression parity; noise semantics unchanged. Workspace: extend convolution_over_sampled.py with a simulate-at-s=2 → fit-at-s=2 round trip (chi² ≈ noise expectation).

Prompt: PyAutoMind/issued/oversampled_psf_simulator.md. Suggested branch: feature/psf-oversample-simulator.

🤖 Generated with Claude Code

[Jammy2211]

Simulator support implemented — at the ship sign-off checkpoint

Five local commits across the task worktree (not pushed): PyAutoArray (+44/−1), PyAutoGalaxy (+105), PyAutoLens (+82/−1), autolens_workspace (simulator.py example), autolens_workspace_test (round-trip test).

Verification

Leg	Result
Round trip	simulate noise-free at s=2 → fit same tracer at s=2: chi² ≈ 1e-30 (unit test in PyAutoLens + workspace-level in convolution_over_sampled.py)
Delta-kernel identity	convolved_padded_image_2d_from padded-frame geometry exact to 1e-14
image_is_convolved	pass-through exact; s=1 behaviour byte-identical (default False)
Suites	PyAutoArray 862, PyAutoGalaxy 945, PyAutoLens 336 (each +1 new test)
Workspace	simulator.py (with the new example section) runs clean; convolution_over_sampled.py all sections pass; regenerated tracked dataset files restored before commit (no binary churn)
Review	5 repos, +281/−2. Verdict: CLEAN
Implementation notes	from_gaussian gained the convolve_over_sample_size pass-through (the example needed it — API completeness); PyAutoLens simulator.py and convolver.py are CRLF files, patched ending-safely

Ship? Order: PyAutoArray → PyAutoGalaxy → PyAutoLens PRs (pending-release), workspace PRs behind the library merges. On approval + merges, the series queue is just phase 4 (docs) and the refactor follow-up.

🤖 Generated with Claude Code

[Jammy2211]

🚢 Shipped (human-approved, with the requested per-project simulate-and-fit coverage): every project now has the oversampled round trip — PyAutoArray test_simulator.py (simulator path vs fit-side convolution, exact), PyAutoGalaxy test_simulate_and_fit_imaging.py (via_galaxies_from, chi²≈0), PyAutoLens test_simulate_and_fit_imaging.py (via_tracer_from, chi²≈1e-30). Final suites: 863 / 946 / 336.

PRs (merge in order): PyAutoArray#359 → PyAutoGalaxy#483 → PyAutoLens#575 → autolens_workspace#233 + autolens_workspace_test#150.

Notes from the final stretch: PyAutoGalaxy's simulator.py turned out to be CRLF and the original patch silently never applied — the Galaxy round-trip test caught it (the Lens test alone had masked it); also purged test-output artifacts that add -A swept from an untracked test_autoarray/output dir (gitignore gap noted for the refactor sweep).

🤖 Generated with Claude Code

[Jammy2211]

✅ Full chain merged in order, each main verified green before proceeding: PyAutoArray#359 → PyAutoGalaxy#483 → PyAutoLens#575 → autolens_workspace#233 + autolens_workspace_test#150. The against-main CI failures resolved exactly as expected once the dependency order was honoured; the navigator catalogue was regenerated in #233. Simulator support for oversampled PSFs is complete, with simulate-and-fit round-trip tests in all three projects.

🤖 Generated with Claude Code