* for z-image fp16 support * cast q_norm and k_norm weight to fp16 for turing gpu * add metirc of int4-fp16 for nunchaku-z-image-turbo-lora
* for z-image fp16 support * cast q_norm and k_norm weight to fp16 for turing gpu * add metirc of int4-fp16 for nunchaku-z-image-turbo-lora