[sinfl] Reduce unnecessary memcpy when memory does not overlap

[yixy-only]

sinfl_copy64 has already ensured that the memory will not overlap before being called, so there is no need to first copy the memory to a temporary buffer before copying it to the target memory.

/* note: src = dst - offs */
if (offs >= 8) {
    /* word copy match */
    sinfl_copy64(&dst, &src);
    sinfl_copy64(&dst, &src);
    do sinfl_copy64(&dst, &src);
    while (dst < out);
}