llama.cpp

c6f6e4f9 - ggml-alloc : fix reuse-parent logic for misaligned sizes (#17884)