sar3d: GPU port of the notebook with four views

The sar3d processor is rewritten as a checked port of Multi_antenna_3D_clean_v1.ipynb:
NVRTC kernels, timing from the trace stamps, notebook parameter names and one cubic
SAR_VOXEL_M. It adds front, side, top and 3D views, drops gpr_cpp and the parity tools,
and the GUI speed now applies only while the socket speed is ignored.

data_acq_and_processing/processing/processors/include/sar3d_processor.hpp

@@ -8,9 +8,11 @@
 
 namespace radar::processing {
 
-// 3D SAR over a synthesized aperture, computed by the external gpr_cpp
-// CUDA/NVRTC library (VNA_AlGo). Emits 2D max-projections of the volume
-// (side XZ and top XY views) as ImageF32 collection payloads.
+// 3D SAR of the reference notebook Multi_antenna_3D_clean_v1.ipynb on an NVIDIA
+// GPU. Publishes max-projections of the normalized volume as ImageF32 payloads:
+// sar3d_front (X-Y), sar3d_side (Z-Y), sar3d_top (X-Z) and sar3d_oblique (the
+// notebook's 3D scene view, axes in the plane of sar3d_oblique_basis), plus
+// sar3d_peak = [level, X, Y, Z] of the volume maximum.
 class Sar3dProcessor final : public ProcessorInterface {
   public:
     Sar3dProcessor();
@@ -25,8 +27,6 @@ class Sar3dProcessor final : public ProcessorInterface {
     ) -> ipc::ResultCollection override;
 
   private:
-    // Pimpl keeps gpr/CUDA headers (cuda.h, nvrtc.h) out of the shared
-    // processor include path; only sar3d_processor.cpp needs them.
     struct State;
     std::unique_ptr<State> state_;
 };