Add end-to-end MXFP4 Metal support

build.rs

@@ -11,6 +11,8 @@ fn main() {
         .include("native/metal")
         .file(metal)
         .flag("-fobjc-arc")
+        .flag("-ffast-math")
+        .flag("-mcpu=native")
         .opt_level(3)
         .compile("ds4_metal");
     println!("cargo:rerun-if-changed=native/media");