refactor(core): embed llama.cpp's server binary directly for LLM inference

refactor(core): embed llama.cpp's server binary directly for LLM inference #2607