Refactor backend to cleaner and more modern API

[getnamo]

Since backend was originally inherited from upstream, along with age in the very rapidly evolving llama.cpp, it would make sense to re-write FLlama to use unreal threading and queue structure along with cleaner and more modern llama.cpp API (see updated examples).

[getnamo]

See e.g.

• main (llama-cli) - https://github.com/ggml-org/llama.cpp/blob/master/examples/main/main.cpp for updated api from CLI app.
• simple-chat - https://github.com/ggml-org/llama.cpp/blob/master/examples/simple-chat/simple-chat.cpp
• save-load-state - https://github.com/ggml-org/llama.cpp/blob/master/examples/save-load-state/save-load-state.cpp for save/load quick context reloading api.