llama.cpp

ggml: allow passing alloc dependencies in graph_optimize

#27301

Merged