diag: resolve kernel at load

app.py

@@ -222,6 +222,8 @@ def load_models() -> str | None:
         if ATTENTION == "vsa":
             VSA_BLOCKS, VSA_GATES = vsa_h3.install(pipe.transformer, sparsity=VSA_SPARSITY)
             print(f"[gen] VSA-H3 on {VSA_BLOCKS} blocks, {VSA_GATES} trained gates", flush=True)
+            # Resolve the sparse kernel now (CPU-side import only) so the log shows the outcome at startup.
+            vsa_h3._resolve_cuda_sparse_op()
         else:
             pipe.transformer.set_attention_backend(ATTENTION)