feat(cnn_v3): add G-buffer visualizer + web sample loader (Phase 7)

C++ GBufViewEffect: renders all 20 feature channels from feat_tex0/feat_tex1
in a 4×5 tiled grid. Custom BGL with WGPUTextureSampleType_Uint; bind group
rebuilt per frame via wgpuRenderPipelineGetBindGroupLayout.

Web tool: "Load sample directory" button — webkitdirectory picker, FULL_PACK_SHADER
compute (matches gbuf_pack.wgsl packing), runFromFeat() skips photo-pack step,
computePSNR() readback + comparison vs target.png side-by-side.

36/36 tests pass. Docs updated: HOWTO.md §9, README, PROJECT_CONTEXT, TODO,
COMPLETED.

handoff(Gemini): CNN v3 Phase 7 done. Next: run train_cnn_v3.py (see HOWTO §3).

1 files changed, 25 insertions, 0 deletions

diff --git a/cnn_v3/src/gbuf_view_effect.h b/cnn_v3/src/gbuf_view_effect.h
new file mode 100644
index 0000000..d4d8139
--- /dev/null
+++ b/cnn_v3/src/gbuf_view_effect.h
@@ -0,0 +1,25 @@
+// GBufViewEffect: Visualizes G-buffer feature textures as a 4×5 channel grid.
+// Inputs: feat_tex0 (rgba32uint, ch 0-7 f16), feat_tex1 (rgba32uint, ch 8-19 unorm8)
+// Output: rgba8unorm tiled channel visualization (downscaled 4× per channel)
+
+#pragma once
+
+#include "gpu/effect.h"
+#include "gpu/sequence.h"
+#include "gpu/wgpu_resource.h"
+
+class GBufViewEffect : public Effect {
+ public:
+  GBufViewEffect(const GpuContext& ctx,
+                 const std::vector<std::string>& inputs,
+                 const std::vector<std::string>& outputs,
+                 float start_time, float end_time);
+
+  void render(WGPUCommandEncoder encoder,
+              const UniformsSequenceParams& params,
+              NodeRegistry& nodes) override;
+
+ private:
+  RenderPipeline pipeline_;
+  BindGroup       bind_group_;
+};