nx_race_native_opt.nx

nx_race_native_opt.nx -- "OURS, OPTIMIZED": the SAME MINSTD LCG as nx_race_native, but with the mod-by-(2^31-1) DIVIDE replaced by MERSENNE STRENGTH-REDUCTION (shifts/adds) -- exactly the optimization gcc -O2 applies automatically. Since 2^31 == 1 (mod 2^31-1), p mod (2^31-1) = fold the 31-bit halves. PURPOSE: prove the measured ~1.7x Nishi-vs-gcc-O2 gap is PURELY auto-strength-reduction (nx_cc doesn't do it yet) -- when the optimization is expressed, Nishi's sovereign codegen reaches the optimizing-gcc frontier. MUST print 668950819 (identical to the divide version) or the reduction is wrong. license_tier: ORIGINAL