added no_mmap & mlock parameters to llama.cpp and removed llamacpp_model_alternative

[axolotlite]

llamacpp users can now specify if they wish to use mmap or mlock
I've removed the llamacpp_model module and renamed llamacpp_module_alternative into it, since the "alternative" module is the default.
I've also changed the default patch size into 512 tokens, which is the current default used by llama-cpp-python library.
There was an issue opened for --no-mmap support #1072

[Priestru]

Thank you! Just what we need to use this one: ggerganov/llama.cpp#1207