perf: cuBLAS matmul fwd/bwd

Route Tensor::matmul and matmul_backward through cuBLAS Sgemm instead of
the hand-written tiled kernel. fp32 → same GEMM up to rounding order, so
the T3 cuBLAS tolerance and downstream grad-checks are preserved.

- cublas.rs: thread-local persistent handle + row-major sgemm helper with
  transpose flags (col-major⟺row-major as the T3 oracle does).
- matmul_backward: dA/dB via cuBLAS OP_T, dropping the two transpose
  kernels + their allocations the T3 version ran.

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

crates/xtrain-cuda/src/ffi.rs

@@ -212,8 +212,9 @@ unsafe extern "C" {
     );
 }
 
-// cuBLAS — used ONLY as a correctness reference for the hand-written GEMM in
-// tests. Declared (and linked, see build.rs) only when CUDA is compiled in.
+// cuBLAS — the production GEMM backend (Phase T7) and the correctness oracle the
+// T3 GEMM tests still compare against. Declared (and linked, see build.rs) only
+// when CUDA is compiled in.
 #[cfg(not(no_cuda))]
 pub type CublasHandle = *mut c_void;
 
@@ -241,3 +242,5 @@ unsafe extern "C" {
 
 #[cfg(not(no_cuda))]
 pub const CUBLAS_OP_N: i32 = 0;
+#[cfg(not(no_cuda))]
+pub const CUBLAS_OP_T: i32 = 1;