lavr: add x86-optimized functions for mixing 1-to-2 s16p with flt coeffs