fix: make plot_array overlays respect imshow_origin "lower" (#565)

[Jammy2211]

Summary

Fixes #565 — community report by @ClarkGuilty, Discussion #14.

With visualize/general.yaml -> imshow_origin: lower, plot_array reflected the image raster about the extent's y midpoint but drew every vector overlay (mask edge, border, origin marker, grid, mesh grid, positions, lines, regions, quiver vectors) in unreflected data coordinates. The figure still looked plausible, so critical curves and caustics silently slid off the arcs they belong to.

The fix follows the pattern _apply_contours already used: two private helpers in autoarray/plot/utils.py reflect overlay y about the raster extent's midpoint when the origin is "lower" (_overlay_yx_for_origin, and _vector_yx_for_origin which also negates dy for quiver rows). plot_array passes every overlay through them before drawing; "upper" is byte-for-byte unchanged. The same audit found the uniform rectangular mesh path of plot_inversion_reconstruction (an imshow honouring the origin) had the identical defect, so _plot_rectangular now returns its raster extent and the lines/regions/grid overlays are reflected there too. Patches are drawn as given and documented as not origin-aware. When extent is None matplotlib draws in pixel-index coordinates under either origin, so no reflection is applied.

The imshow_origin config comment now records the caveat from the report: the setting is presentation only, and under lower the displayed y axis and reported model y parameters disagree in sign.

API Changes

None. Both helpers are private; the config key, its default and every public plot_* signature are unchanged. User-visible behaviour change: with imshow_origin: lower, overlays now land on the image features they describe.

Test Plan

• New test_autoarray/plot/test_array.py: three origin-agnostic regression tests (positions marker, asymmetric polyline, uniform-mesh inversion grid), each parametrised over upper / lower. They read the drawn overlay back from the axes, map it through the image's own extent and origin, and assert the array is bright there — so they keep passing under any correct fix rather than pinning today's layout. Red on unfixed main for every lower case (assert 0.0 == 1.0), green now.
• python -m pytest test_autoarray/plot/ — 41 passed
• python -m pytest test_autoarray/ — 1981 passed, 4 xfailed
• CI matrix (3.12 / 3.13)

Known remaining lower quirk, out of scope: zoom_to_brightest in plot_inversion_reconstruction sets axis limits from an unreflected zoom extent, so on a uniform mesh under lower the zoom window can sit over the wrong part of the image. Pre-existing; noted on #565.

Full API Changes (for automation & release notes)

Added

• autoarray.plot.utils._overlay_yx_for_origin(yx, extent, origin_imshow) (private)
• autoarray.plot.utils._vector_yx_for_origin(vector_yx, extent, origin_imshow) (private)
• autoarray.plot.inversion._plot_rectangular now returns the raster extent (uniform path) or None (private)

Removed

• none

Migration

• none

Generated by the PyAutoLabs agent workflow.

🤖 Generated with Claude Code