refactor: k×s surface — unify the evaluate/pre-bin/convolve sequence

[Jammy2211]

Proper Refactor Agent run on the k×s surface (PyAutoMind/refactor/autoarray/kxs_surface_refactor.md; RefactorDecision: too-large score 10 → conservative cut, invariant = suites unchanged, witnesses autoarray+autogalaxy, API guard none-expected, cap safe). Prime candidate: operate/image.py's three blurred-image variants each re-implement evaluate → maybe-pre-bin → convolve-or-delegate; one shared core collapses the three branches. PyAutoArray pieces (util, mapper pair, validation helper) assessed in-place — left alone if clean, recorded either way. convolver/ split stays out of scope per #360.

🤖 Generated with Claude Code

[Jammy2211]

🚢 Shipped: PR #489 (pending-release). Assessment record for the PyAutoArray surface (no edits): convolve_bin_segment_ids_from/binned_to_convolve_size_from are single-purpose with the xp split intentionally mirroring OverSampler.binned_array_2d_from; the mapper property/method pair is minimal (property = k=1 case); the divisibility validator is one predicate. Consolidating any of them would add indirection without removing duplication. The convolver/ package split remains out of scope per PyAutoArray#360. Witness note: the workspace script ran from origin/main's content — the canonical checkout is on another session's live branch (fix/visualization-jit-test-mode-output-path), untouched.

🤖 Generated with Claude Code

[Jammy2211]

✅ Merged as PR #489 (main green). The k×s refactor debt is repaid agent-first; closing.

🤖 Generated with Claude Code