nx_p256_solinas_fast.nx

nx_p256_solinas_fast.nx -- a FASTER NIST P-256 Solinas reducer, bit-exact with the production _p256_solinas_reduce. The production version issues ~60 _r_addw FUNCTION CALLS, each propagating a carry up a 9-limb memory accumulator on every word add. This version DEFERS the carry: it sums each limb as a wide 64-bit value (max ~6*(2^32-1) < 2^35, no overflow) with plain array adds (no calls, no per-add carry loop), then does ONE normalization pass before the identical reduce/subtract tail. Same S-term table, same _r_reduce9 / _r_cmp8 helpers, same final POS-NEG mod p -> the result is identical (proven by nx_p256_solinas_fast_difftest). Uses NO compiler intrinsic, so it builds with the live compiler; kept SEPARATE so the production reducer is untouched until measured+greenlit. license_tier: ORIGINAL