cuda-core: require pytorch-gpu in examples env

Use the explicit `pytorch-gpu` metapackage for the examples environment so PyTorch resolves to a CUDA-enabled build by design.

Made-with: Cursor

Author: cpcloud

cuda_core/pixi.toml

@@ -26,7 +26,6 @@ pyglet = "*"
 
 [feature.examples.target.linux.dependencies]
 cupy = "*"
-pytorch = "*"
 pytorch-gpu = "*"
 libgl-devel = "*"
 gxx = "*"