Add installation instructions & llama.cpp as a submodule

[HaroldBenoit]

Hello, thank you very much for this toolkit. This is very useful for the community, and this PR aims to make a few improvements.

• It adds an explicit requirements.txt & installation instructions in the README.md.

  • The current dependencies specified in the README.md don't work out of box with pip or uv.

• It adds explicitly llama.cpp as a submodule.

  • This allows a clear way to obtain llama binaries such as llama-quantize necessary for scripts such as quant/gguf/run_quant.sh .
  • This also fixes import errors (e.g. MistralTokenizerType is not importable) when running quant/gptq/pack_gptq_into_gguf.py, as the pypi version of gguf is not up-to-date, and llama.cpp implictly requires to point to the local upstream version stored in the gguf-py folder.