nx_superopt_emit.nx

nx_superopt_emit.nx -- emit COMPETITIVE x86-64 machine code for a superoptimized program (the slot/op/a/b form from nx_superopt_opt). This is the team building UP from machine code: SYNTHESIZE the optimal op sequence, then EMIT it as tight assembly to race gcc -O2. Tightness comes from the same register-allocation insight proved in nx_regalloc_calls: keep the running result in one register (reuse a dead operand's register in place), and use `lea` to fuse a shift/add with its move so each op is ~1 instruction -- exactly how gcc's strength reduction emits. Input x in %rdi, result in %rax. license_tier: ORIGINAL op/a/b: op[t] in {ADD,SUB,MUL,SHL,SAR,XOR,AND,OR}; a[t],b[t] are slot indices (0=x, 1=const0, 2=const1, t+3=result of op t); for shifts b[t] is the shift amount.