Mlipy

Added:
- server/client: `split_mode`, `tensor_split`, `main_gpu`.

Changed:
- README: New models.

Changed:
- README: instructions how to build `llama.cpp`
- README: New models.

Changed:
- README: instructions how to build `llama.cpp` and `candle`.
- examples: more examples and improvements in `sync_demo.py`.
- server: fixed format output for `messages_syntax='llama'`.

Fixed:
- server: no need to parse printed/echoed prompt thanks to new instructions how to build `llama.cpp` and `candle`.

Changed:
- server: improved `messages_syntax` formatting.

Fixed:
- server: better handling of `stop` words.

Fixed:
- server: optional `messages_syntax`.

Changed:
- server: support for `messages_syntax` formating `chatml`, `llama`, `zephyr`.

Fixed:
- client: fixed `llama.cpp` examples, `ctx_size`.