build: package universal Windows runtime and verify automatic CPU fallback

tools/embeddinggemma2/build_runtime.py

@@ -22,6 +22,11 @@ def main():
     args = parser.parse_args()
     if not 1 <= args.jobs <= 16:
         parser.error("jobs must be 1–16")
+    vulkan_loader = None
+    if os.name == "nt" and args.backend in {"auto", "vulkan"}:
+        from provision_vulkan import provision
+        sdk, vulkan_loader = provision()
+        os.environ["VULKAN_SDK"] = str(sdk)
     source = ROOT / ".searchi/references/llama.cpp"
     build = ROOT / ".searchi/runtime" / args.backend
     if not source.exists():
@@ -51,6 +56,9 @@ def main():
          "-DCMAKE_BUILD_TYPE=Release", f"-DLLAMA_CPP_DIR={source}",
          f"-DSEARCHI_ACCELERATION={args.backend}"])
     run([cmake_exe, "--build", build, "--config", "Release", "--parallel", args.jobs])
+    if vulkan_loader:
+        for name in ["vulkan-1.dll", "VULKAN-LICENSE.txt"]:
+            shutil.copy2(vulkan_loader / name, build / "Release" / name)
     print(f"Runtime build complete: {build}")