Mlipy

Changed:
- Updated `langchain`, `huggingface-hub`.

Added:
- `llama.cpp`: support for both `main` and `llama-cli` executable.

Added:
- `llama.cpp` supports: `flash_attn`, `grp_attn_n`, `grp_attn_w`

Fixed:
- Memory leak fixed by killing/terminating `llama.cpp` process called `main`.

Fixed:
- Stop word detection on slow stream reads.

Fixed:
- Stop word detection.
- `stderr` has max `1.0` second to finish after process is killed/terminated.
- Fixed exception `KeyError: 'prompt'` in `prompt: str = msg['prompt']`.
- Fixed killing subprocess with its child process.