* ggml : add predefined list of CPU backend variants to build * update CPU dockerfiles
ggml : automatic selection of best CPU backend (llama/10606)