autograd: tape engine + grad accumulation

Var = Rc<RefCell<VarNode>> on a define-by-run tape: value + optional grad +
parents + backward closure. backward() seeds a scalar loss, walks reverse
topo order, and pushes grads to parents. push_grad always SUMs into the grad
slot — the fan-out accumulation path T3 lacked. Per-crate build.rs emits the
no_cuda cfg (does not propagate); engine gated, grad_check stays host-only.

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

crates/xtrain-autodiff/build.rs

@@ -0,0 +1,27 @@
+use std::env;
+use std::path::Path;
+use std::process::Command;
+
+// xtrain-autodiff's `Var` tape calls GPU ops (via xtrain-tensor), so it gates
+// those call sites behind `not(no_cuda)` — the same per-crate convention as
+// xtrain-cuda / xtrain-tensor (cfg does not propagate across crates). The
+// grad_check harness itself is host-only and always compiles. This script only
+// detects nvcc and emits the cfg; it compiles no CUDA.
+fn main() {
+    println!("cargo:rustc-check-cfg=cfg(no_cuda)");
+
+    let cuda_path = env::var("CUDA_HOME")
+        .or_else(|_| env::var("CUDA_PATH"))
+        .unwrap_or_else(|_| "/usr/local/cuda".to_string());
+
+    if !nvcc_available(&cuda_path) {
+        println!("cargo:rustc-cfg=no_cuda");
+    }
+}
+
+fn nvcc_available(cuda_path: &str) -> bool {
+    if Command::new("nvcc").arg("--version").output().is_ok() {
+        return true;
+    }
+    Path::new(&format!("{cuda_path}/bin/nvcc")).exists()
+}