gemm: tiled F32 forward + transpose + backward (dA/dB)

Hand-written tiled GEMM (csrc/ops/gemm.cu, TILE_SIZE=32, FP32 accumulate,
boundary-masked) plus an out-of-place transpose kernel. Wire both through
xtrain-cuda FFI (no_cuda-gated) and expose at the tensor level:
Tensor::matmul, transpose_2d, and matmul_backward computing
dA = dC·Bᵀ and dB = Aᵀ·dC by materializing transposes and reusing the
forward. Also declare cuBLAS sgemm FFI + link cublas, used only as a
correctness reference in tests.

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

crates/xtrain-cuda/build.rs

@@ -22,6 +22,8 @@ fn main() {
     println!("cargo:rustc-link-search=native={cuda_path}/lib64");
     println!("cargo:rustc-link-lib=dylib=cudart");
     println!("cargo:rustc-link-lib=dylib=cuda");
+    // cuBLAS is used only as a correctness reference for the hand-written GEMM.
+    println!("cargo:rustc-link-lib=dylib=cublas");
 
     cc::Build::new()
         .cuda(true)
@@ -29,6 +31,7 @@ fn main() {
         .flag("-gencode=arch=compute_120,code=sm_120")
         .file("../../csrc/test/vecadd.cu")
         .file("../../csrc/ops/elementwise.cu")
+        .file("../../csrc/ops/gemm.cu")
         .compile("xtrain_cuda_kernels");
 }