nx_rsa2048_mont.nx

nx_rsa2048_mont.nx -- Montgomery modular multiplication + modexp for the 2048-bit RSA modulus. Replaces the O(bits*n) bit-by-bit reduction (nx_rsa2048_mod.nx, "a few seconds per verify") with CIOS Montgomery multiplication (O(n^2) limb ops), measured to cut TLS cert-verify from ~1.5s to a few ms (sovereign native perf timers in nx_https_get / _session_run). 32-bit limbs in 64-bit words make CIOS overflow-safe: the inner accumulate t[j] + a[j]*b[i] + carry has max value (2^32-1)^2 + 2*(2^32-1) = 2^64 - 1, which fits one unsigned 64-bit word exactly. Carry/low are extracted with (x>>32)&MASK / x&MASK (correct under arithmetic shift+mask). Side channels: this verifies PUBLIC server certificates (modulus, signature, and exponent are all public), so constant-time is not required here. Correctness is pinned by KAT equality against the bit-by-bit rsa2048_mod_exp in nx_rsa2048_mont_gate.nx -- both must produce identical s^e mod n. license_tier: INDEPENDENT_REDERIVE genealogy_id: international-research-sources/ietf/rfc_8017 (Montgomery 1985; Koc CIOS) lineage_id: nishi_rsa2048_mont_q10