nx_alu_divider_r8.nx

nx_alu_divider_r8.nx -- radix-8 divider (3 quotient bits/iteration => W/3 subtract-stages). Bottom-up latency iteration on the divider car (don't move up until S-class): radix-2 (W) -> radix-4 (W/2) -> radix-8 (W/3) -> ... toward the multiplicative frontier (Goldschmidt ~log W). Same proven digit-recurrence invariant, q-digit in {0..7} via 7 compare-multiples + a cascade. Reuses nx_alu_divider_r4 (radix-2 + radix-4) helpers. width must be a multiple of 3. license_tier: ORIGINAL