Llama-cpp-cffi

Added:
- `llava` high-level API calls
- `minicpmv` high-level API support

Changed:
- Updated `llama.cpp`.

Changed:
- Updated `llama.cpp`.

Changed:
- Updated `llama.cpp`.

Changed:
- Updated `llama.cpp` with **RWKV 6** support.

Changed:
- Updated `llama.cpp` with **RWKV** support.

Added:
- `LLAMA_CPP_BACKEND` which can be `cuda`, `vulkan` or `cpu`.

Changed:
- Updated `llama.cpp`.
- Updated requirements.
- CUDA backend imports only 12.6 library.

Fixed:
- `Options.top_p` check using `isinstance`.