Mlipy

Added:
- Server: download `GGUF` model if it does not exist using `huggingface-hub`.
- Examples: `candle`-based sync examples `codellama`, `llama`, `mistral`.

Changed:
- Instructions how to pre-download `GGUF` models required for `llama.cpp`.

Deprecated:
- Server: `GGUF` model path is not passed anymore.

Fixed:
- Server: engine `candle`, kind `llama`.
- Server: engine `candle`, kind `mistral`.

Added:
- Examples: sync_demo uses candle/llama.cpp for llama2/mistral/stablelm

Changed:
- Server does not assert anymore model names/ids.

Added:
- Install instructions for Debian, Ubuntu, ArchLinux, Manjaro, macOS.
- Server: LlamaCppParams, CandleParams, LLMParams.
- Examples: sync_demo.py, async_demo.py, langchain_sync_demo.py, langchain_async_demo.py

Changed:
- Client examples.
- Server: data/msg of type LLMParams.
- Examples use context size 512.
- Insomnia examples use context size 1024.
- Package `mlipy` renamed to `mli`.
- Streaming sync and async langchain functions.
- Less noisy output on token generation.

Added:
- Park first version.