llama.cpp

Fix Qwen3-Embedding pre-tokenizer hash

#15030

Merged