Now that there is an SSE and an AVX implementation,
there is no need to have an scalar implementation for Chacha20.