feat(cnn_v3): Phase 4 complete — CNNv3Effect C++ + FiLM uniform upload

- cnn_v3/src/cnn_v3_effect.{h,cc}: full Effect subclass with 5 compute
  passes (enc0→enc1→bottleneck→dec1→dec0), shared weights storage buffer,
  per-pass uniform buffers, set_film_params() API
- Fixed WGSL/C++ struct alignment: vec3u has align=16, so CnnV3Params4ch
  is 64 bytes and CnnV3ParamsEnc1 is 96 bytes (not 48/80)
- Weight offsets computed as explicit formulas (e.g. 20*4*9+4) for clarity
- Registered in CMake, shaders.h/cc, demo_effects.h, test_demo_effects.cc
- 35/35 tests pass

handoff(Gemini): CNN v3 Phase 5 next — parity validation (Python ref vs WGSL)

1 files changed, 5 insertions, 0 deletions

diff --git a/src/tests/gpu/test_demo_effects.cc b/src/tests/gpu/test_demo_effects.cc
index c2588f2..f5af5a9 100644
--- a/src/tests/gpu/test_demo_effects.cc
+++ b/src/tests/gpu/test_demo_effects.cc
@@ -84,6 +84,11 @@ static void test_effects() {
            fixture.ctx(), std::vector<std::string>{"source"},
            std::vector<std::string>{"gbuf_feat0", "gbuf_feat1"}, 0.0f,
            1000.0f)},
+      {"CNNv3Effect",
+       std::make_shared<CNNv3Effect>(
+           fixture.ctx(),
+           std::vector<std::string>{"gbuf_feat0", "gbuf_feat1"},
+           std::vector<std::string>{"cnn_v3_output"}, 0.0f, 1000.0f)},
   };
 
   int passed = 0;