optimize Dockerfile by using precompiled wheels index and removing compiler tools

mock to realtime local model execution with llama.cpp