llama.cpp

kv-cache : fix seq_rm with seq_id == -1

#15226

Merged