* x86_64/aes-decrypt-internal.asm: Rearrange register allocation.

Put SA--SD in %eax--%edx, so the second byte can be accessed as
%ah-%dh. TD is not needed, SD can be reused. Use the register that
is saved for the outer loop counter, getting it off the stack.
* x86_64/aes-encrypt-internal.asm: Likewise.

* x86_64/aes.m4 (HREG, MOVE_HREG): New macros.
(XREG): Fixed bug in handling of %r8 and %r9.
(AES_ROUND): Use MOVE_HREG.

Rev: nettle/x86_64/aes-decrypt-internal.asm:1.2
Rev: nettle/x86_64/aes-encrypt-internal.asm:1.9
Rev: nettle/x86_64/aes.m4:1.6