build(crt-train): register output_smoothness.cu in ml-alpha build.rs KERNELS

crates/ml-alpha/build.rs

@@ -21,12 +21,10 @@ const KERNELS: &[&str] = &[
     "variable_selection",            // Phase 2D: TFT-style per-feature gating
     "attention_pool",                // Phase 3: single-Q learned content summary at CfC k=0
     "reduce_axis0",                  // Phase B: cross-batch param-grad reducer
+    "output_smoothness",             // CRT.train: per-horizon adjacent-position prob-jitter penalty
 ];
 
-// Cache bust v11 (2026-05-17): K-loop parallelization Phase B —
-// new reduce_axis0.cu kernel + block-per-batch refactor of
-// cfc_step_batched (fwd+bwd). Old cubins don't have the new symbols.
-// Force fresh nvcc compile.
+// Cache bust v12 (2026-05-20): output_smoothness.cu added — CRT.train intervention A.
 
 fn main() {
     println!("cargo:rerun-if-changed=build.rs");