llama.cpp

server : fix processing of multiple back-to-back mtmd chunks

#21107

Merged