tensor: add scale elementwise CUDA kernel + FFI

New csrc/ops/elementwise.cu (out[i]=in[i]*alpha), compiled by
xtrain-cuda/build.rs and exposed via launch_scale_f32 FFI, gated behind
not(no_cuda) like the existing vecadd smoke test.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

crates/xtrain-cuda/build.rs

@@ -28,6 +28,7 @@ fn main() {
         .cudart("shared")
         .flag("-gencode=arch=compute_120,code=sm_120")
         .file("../../csrc/test/vecadd.cu")
+        .file("../../csrc/ops/elementwise.cu")
         .compile("xtrain_cuda_kernels");
 }