Add optika.wavefields: physical-optics propagation through a SequentialSystem

[jacobdparker]

Summary

Stacked on #172 (which is stacked on #171; will retarget as they merge). The aperture-bounds fix in #171 is required for correct sampling of apertures with internal transformations (covered by a regression test here).

Adds a physical-optics path that mirrors the geometric raytrace: the complex scalar field is sampled on every surface with stratified random samples and propagated surface-by-surface with the Rayleigh-Sommerfeld (RS-I) diffraction integral, producing Huygens PSFs on the sensor.

• optika.wavefields: WavefieldVectorArray (complex amplitude, position, surface normal, sample area; transformation dunders mirror RayVectorArray) and rayleigh_sommerfeld(), a chunked, numexpr-fused kernel whose peak memory is bounded by chunk_size x N_src x 16 B instead of the full N_src x N_dst outer product.
• Surface.wavefield_samples / propagate_wavefield: stratified random sampling of each aperture (via na.Cartesian2dVectorStratifiedRandomSpace, which converts aliasing into broadband noise), with an optional per-surface bound to window surfaces near an intermediate focus.
• Diffraction gratings: the ruling phase 2 pi m N(r) is applied at each sample, with the groove-counting function N(r) computed as a Gauss-Legendre line integral of the local groove density - exact for polynomial (VLS) ruling spacings and consistent with the sign convention of optika.rulings.incident_effective.
• SequentialSystem.wavefield / psf: orchestration over the surface chain, plane- or spherical-wave initial conditions from the system object, and a sensor grid auto-centered on the geometric image centroid.

The docstring Notes document the key sampling constraint: the brute-force RS sum is only well-sampled between near-conjugate surfaces; flat masks in collimated space are better folded into the pupil than treated as propagation steps.

Validation

• Airy pattern of an ideal paraboloid: first zero at 1.22 lambda f/D, null depth ~3e-6.
• Chunk invariance (bitwise) and energy conservation to 0.03%.
• Huygens vs geometric centroids agree through a 90-degree Newtonian fold and a concave VLS grating (~20 nm).
• End-to-end on the ESIS f1 spectrograph at O V 630 A: PSF FWHM matches the no-free-parameter lambda L/D prediction of the channel pupil in both axes; image centroids match the raytrace to <0.25 um across the FOV and to <0.6 um across an 18.5 mm dispersion span (He I 584 to O V 630).

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 91.01498% with 54 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.64%. Comparing base (8455e8d) to head (60267ad).

Files with missing lines	Patch %	Lines
optika/systems/_sequential.py	79.80%	21 Missing ⚠️
optika/wavefields/_wavefield_vectors.py	82.02%	16 Missing ⚠️
optika/surfaces.py	82.35%	15 Missing ⚠️
optika/systems/_wavefield_test.py	99.25%	1 Missing ⚠️
optika/wavefields/_wavefield_functions.py	92.30%	1 Missing ⚠️

Additional details and impacted files

@@                   Coverage Diff                   @@
##           feature/zernike-sag     #173      +/-   ##
=======================================================
- Coverage                99.37%   98.64%   -0.74%     
=======================================================
  Files                      120      127       +7     
  Lines                     6248     6849     +601     
=======================================================
+ Hits                      6209     6756     +547     
- Misses                      39       93      +54     

Flag	Coverage Δ
unittests	98.64% <91.01%> (-0.74%)	⬇️

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.