QNN Module API(new frontend) Preliminary Support

[oreomaker]

!!!Do not merge until QNN backend qwen is runable!!!

Support assigning backends for modules and setup & executing computing graphs.
Better design for module backend splitting and 'tensor.to()' should be discussed.

TODO:

• KVCacheNPU create and reuse
• shadow execute used in QNN model
• QNN tensor scale loading and setting

[yirongjie]

1. add Layer::use_layername_2_tensorname to select whether use layername_2_tensorname;
2. add to function GetBackendCreator in file Backend.cpp;

    if (type == MLLM_QNN) {
        Layer::use_layername_2_tensorname = false;
    }

3. remove :

#ifdef USE_QNN
       // when use QNN, tensors are not reused
       // TODO: tensor reuse and pipeline support
       layername_2_tensorname[layer_next_name] = layer_next_name;
#else