llama : add stdexcept header for std::runtime_error

[danbev]

This commit adds the stdexcept header to llama-arch.cpp to allow the use of std::runtime_error.
Currently the following compilation is generated:

/work/ai/new-vision-api/src/llama-arch.cpp: In member function‘std::string BASE_TN_IMPL<Tname, Ttensor>::str() const [with Tname = llm_arch; Ttensor = llm_tensor; std::string = std::__cxx11::basic_string<char>]’:
/home/danbev/work/ai/new-vision-api/src/llama-arch.cpp:1544:20: error: ‘runtime_error’ is not a member of ‘std’
 1544 |         throw std::runtime_error(format("Cannot find tensor name mapping for arch %d", arch));
      |                    ^~~~~~~~~~~~~