Fixed aperture bounding boxes ignoring the internal transformation

[jacobdparker]

Summary

AbstractPolygonalAperture.bound_lower/bound_upper were computed from the raw vertices, while __call__ and wire() apply the aperture's internal transformation. For apertures with a nonzero internal transformation (e.g. a decentered grating aperture), every consumer of the bounding box (stratified sampling grids, sensor pixel grids, pupil normalization) saw the wrong region of the surface.

Found while validating a physical-optics model of ESIS f1: the grating's trapezoidal aperture is decentered by -12.26 mm, so the sampling grid only overlapped a ~4.7 mm sliver of the real 16.9 mm aperture, broadening the Huygens PSF by 4x in the dispersion direction.

Two related defects fixed in the same commit:

• EllipticalAperture bounds transformed only the bounding-box corner points, which is incorrect under rotation; its bounds are now derived from the wire, matching CircularSectorAperture.
• RectangularAperture.__call__ expressed the rectangle in terms of bound_lower/bound_upper while also inverse-transforming the position, which double-counts the internal transformation once the bounds include it; it now tests against half_width directly.

Tests

The generic bound tests now assert that the bounds enclose the wire, and that the wire centroid is accepted by active, non-inverted apertures; plus a targeted regression test for a decentered RectangularAperture. Comparisons use nominal values (uncertain parameters may be redrawn between independent evaluations under pytest-randomly) with a small tolerance for wire-interpolation round-off.

pytest optika/apertures: 20485 passed.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.35%. Comparing base (e9a4eb5) to head (9479771).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #171   +/-   ##
=======================================
  Coverage   99.34%   99.35%           
=======================================
  Files         116      116           
  Lines        5974     6024   +50     
=======================================
+ Hits         5935     5985   +50     
  Misses         39       39           

Flag	Coverage Δ
unittests	99.35% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[roytsmart]

My concern is that if the aperture is circular/elliptical, taking the min/max of the wore might not capture the full extent. Can we handle these cases analytically?

[jacobdparker]

Good point — switched these to a closed form rather than sampling the wire. A boundary point is c + a*e_a*cos(t) + b*e_b*sin(t), so the extent along each world axis is sqrt((a*e_a)^2 + (b*e_b)^2) (since max(A cos t + B sin t) = sqrt(A^2 + B^2)). That's exact under rotation and reduces to the (radius.x, radius.y) box when untransformed. CircularAperture is already analytic and rotation-invariant, so it's unchanged. Done in 9479771.