* llama : improve infill support ggml-ci * llama : add more FIM token strings ggml-ci * server : update prompt on slot restore (#9800) * gguf : deprecate old FIM token KVs
json