bench: add --tensor-read-lazy (#27881)

* bench: add --tensor-read-lazy

* rm the alias

* rename to LLAMA_LAZY_MODE_*

common/common.h

@@ -483,7 +483,7 @@ struct common_params {
     enum llama_split_mode split_mode = LLAMA_SPLIT_MODE_LAYER; // how to split the model across GPUs
     enum llama_load_mode  load_mode  = LLAMA_LOAD_MODE_AUTO; // how to load the model
 
-    enum llama_tensor_read_lazy tensor_read_lazy = LLAMA_TENSOR_READ_LAZY_AUTO; // on-demand reading of tensors marked by the arch
+    enum llama_lazy_mode lazy_mode = LLAMA_LAZY_MODE_AUTO; // on-demand reading of tensors marked by the arch
 
     common_cpu_params cpuparams;
     common_cpu_params cpuparams_batch;