refactor(cnn_v3): GBufferEffect cleanup

Remove dead code and reduce duplication:
- drop create_bilinear_sampler() (never called)
- drop update_pack_bind_group() stub and pack_bind_group_ member
- drop node_feat0_/node_feat1_; use output_nodes_[0/1] directly
- Compose({}, src) consistently for all three pipelines
- extract clear_r8_node() helper to replace two identical 10-line blocks

No behavior change. 36/36 tests pass.

1 files changed, 0 insertions, 4 deletions

diff --git a/cnn_v3/src/gbuffer_effect.h b/cnn_v3/src/gbuffer_effect.h
index c39219b..13d394d 100644
--- a/cnn_v3/src/gbuffer_effect.h
+++ b/cnn_v3/src/gbuffer_effect.h
@@ -68,8 +68,6 @@ class GBufferEffect : public Effect {
   std::string node_depth_;
   std::string node_shadow_;
   std::string node_transp_;
-  std::string node_feat0_;
-  std::string node_feat1_;
 
   // Owned scene and camera — populated by set_scene()
   Scene  scene_;
@@ -88,7 +86,6 @@ class GBufferEffect : public Effect {
 
   // Pass 4: Pack compute pipeline
   ComputePipeline  pack_pipeline_;
-  BindGroup        pack_bind_group_;
   UniformBuffer<GBufResUniforms>    pack_res_uniform_;
   UniformBuffer<GBufLightsUniforms> lights_uniform_;
 
@@ -102,7 +99,6 @@ class GBufferEffect : public Effect {
   void create_pack_pipeline();
 
   void update_raster_bind_group(NodeRegistry& nodes);
-  void update_pack_bind_group(NodeRegistry& nodes);
 
   void upload_scene_data(const Scene& scene, const Camera& camera,
                          float time);