sync llama.cpp (for overflow fix)

In [llama.cpp 26a48ad6](https://github.com/ggml-org/llama.cpp/commit/26a48ad699d50b6268900062661bd22f3e792579), an integer overflow was fixed. Could we have the change sync'ed back into ggml and (I guess whisper.cpp)?

This is useful where the libraries are built separately through LLAMA_USE_SYSTEM_GGML=ON.