x86-asm: Fix lar opcode operands

lar can accept multiple sizes as well (wlx), like lsl.  When using
autosize it's important to look at the destination operand first;
when it's a register that one determines the size, not the input
operand.

tests/asmtest.S

@@ -625,8 +625,15 @@ int $0x10
     clflush 0x1000(%rax,%rcx)
     fxsaveq (%rdx)
     fxrstorq (%rcx)
+
 #endif
 
+    lar %ax,%dx
+    lar %eax,%dx
+    lar %ax,%edx
+    lar %eax,%edx
+    lar %ax,%rdx
+    lar %eax,%rdx
     emms
     movd %edx, %mm3
     movd 0x1000, %mm2