lib: compress ZUC state for SSE/AVX/AVX2

To improve cache utilization, ZUC state is compressed as it
was before for SSE/AVX/AVX2, so all the state is in contiguous
memory, instead of being scattered due to some lanes not being
used (SSE/AVX use 4 lanes, AVX2 uses 8 lanes, while AVX512 uses 16 lanes).