Make llama.cpp depends on ggml

[zenixls2]

• cmake: ExternalProject

• Makefile git clone

• changes required in ggml:

  • move examples/common* out to include/ggml/
  • move some frequently used functions in llama.cpp to include/ggml/llm and src/

• changes required in llama.cpp:

  • simplify llama.cpp by including/extending ggml/include/ggml/llm/
  • CMakeFile to re-export flags from ggml

Don't want to depend on conan since it adds more dependencies. And I think it's time to make decisions... as the project's getting more and more complex.

[marella]

I was also thinking of suggesting something similar. For example, llama_sample_* functions can be reused by other models.

Author might already have similar plans https://github.com/ggerganov/ggml/pull/145#issuecomment-1544733902