reg_id_t width check broken by OPSZ_NA=0

I noticed that 053f6b6a PR #3497 broke this check:

     * The reg_id_t type is now wider, but for x86 we ensure our values
     * all fit via an assert in d_r_arch_init().

That PR set OPSZ_NA to 0, which broke the check, which I believe is this:

    IF_X86(ASSERT(CHECK_TRUNCATE_TYPE_byte(OPSZ_LAST)));