T1: scaffold repo + Rust/CUDA build chain (vecadd smoke test)

Stand up the xtrain project skeleton: a Cargo workspace mirroring xserv's
csrc/ + crates/ layout, with a single xtrain-cuda crate that wraps the CUDA
Runtime over hand-written extern "C" FFI. build.rs compiles csrc/test/vecadd.cu
via the cc crate targeting sm_120 (RTX 5090) and links cudart.

A gated integration test runs the vector-add kernel on the GPU and asserts the
result. When nvcc is absent (local GPU-less machine), build.rs skips CUDA
compilation and sets a `no_cuda` cfg so host-side cargo check still works.

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

crates/xtrain-cuda/src/ffi.rs

@@ -0,0 +1,32 @@
+use std::ffi::c_void;
+use std::os::raw::c_char;
+
+pub type CudaStream = *mut c_void;
+
+pub const CUDA_MEMCPY_H2D: i32 = 1;
+pub const CUDA_MEMCPY_D2H: i32 = 2;
+
+pub const CUDA_SUCCESS: i32 = 0;
+pub const CUDA_ERROR_OUT_OF_MEMORY: i32 = 2;
+
+unsafe extern "C" {
+    // --- Device ---
+    pub fn cudaGetDeviceCount(count: *mut i32) -> i32;
+    pub fn cudaSetDevice(device: i32) -> i32;
+    pub fn cudaDeviceSynchronize() -> i32;
+
+    // --- Memory ---
+    pub fn cudaMalloc(devptr: *mut *mut u8, size: usize) -> i32;
+    pub fn cudaFree(devptr: *mut u8) -> i32;
+    pub fn cudaMemcpy(dst: *mut u8, src: *const u8, count: usize, kind: i32) -> i32;
+
+    // --- Error ---
+    pub fn cudaGetErrorString(error: i32) -> *const c_char;
+}
+
+// The vector-add smoke-test kernel, compiled from csrc/test/vecadd.cu by build.rs.
+// Only linked when CUDA is actually compiled (i.e. nvcc was present).
+#[cfg(not(no_cuda))]
+unsafe extern "C" {
+    pub fn launch_vecadd_f32(a: *const f32, b: *const f32, c: *mut f32, n: i32, stream: CudaStream);
+}