Record that the branch is based on feature/cuda_clusterfinder, point
step 6 at the CMake and test setup it inherits, and drop stale loose
ends.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GPU version of PixelHistogram: one histogram per pixel, filled on the
device from batches of frames.
- PixelHistogramCUDA: copy/compute stream pipeline with two device slots
and per-slot copy_done/kernel_done events; bin-major uint32_t
accumulator kept on the GPU; values() and values_pixel() return the
CPU classes' pixel-major StorageType layout, with counts capped once.
- Kernels: fill_bin_major (one thread per pixel, no atomics) and
bin_to_pixel_major (transpose + cap).
- RAII_device.hpp: add HostRegistration, AsyncDeviceBuffer and Event.
- docs/dev: design and progress guide.
Not yet wired into CMake and not tested end to end; values() and
values_pixel() are verified only with standalone nvcc tests.
