feat(arch): scale and exponent-extract the scalar FP16 in the extension layer

Assisted-by: GLM 5.3 Flash

asm/extension_amd64_test.go

@@ -35,6 +35,8 @@ func TestAmd64ExtensionRegistry(t *testing.T) {
 		{"VCVTSH2SI", 2},
 		{"VADDPH", 1},
 		{"VSQRTPH", 1},
+		{"VSCALEFSH", 1},
+		{"VGETEXPSH", 1},
 	} {
 		cands, ok := LookupExtension(arch.AMD64, tt.mnem)
 		if !ok {
@@ -48,8 +50,8 @@ func TestAmd64ExtensionRegistry(t *testing.T) {
 			t.Errorf("the %s lookup is not case-insensitive", tt.mnem)
 		}
 	}
-	if got := arch.Extensions(arch.AMD64); len(got) != 46 {
-		t.Errorf("the amd64 layer registers %d instructions, want 46", len(got))
+	if got := arch.Extensions(arch.AMD64); len(got) != 48 {
+		t.Errorf("the amd64 layer registers %d instructions, want 48", len(got))
 	}
 	if _, ok := LookupExtension(arch.AMD64, "NOSUCHINSTR"); ok {
 		t.Error("a non-extended mnemonic resolved")