Optics: parallelize PSF cache construction

psf_lookup_optimization_plan.md

@@ -116,7 +116,8 @@ case.  The direct evaluator remains a selectable regression reference.
 Implemented as one immutable `PsfKernelCache` per invocation, built after CLI
 parsing and passed read-only through the existing frame splat workers.  The
 cache uses 64 by 64 phase intervals (65 nodes on each axis), 4-point
-Gauss-Legendre pixel-area quadrature, and bilinear phase interpolation.  The
+Gauss-Legendre pixel-area quadrature, bilinear phase interpolation, and a
+static OpenMP phase-plane construction loop.  The
 `--psf-direct` regression mode uses an independent 8-point quadrature
 reference; events whose requested support exceeds the cache use that path
 automatically.