Add the ability to manage settings located with the model.

[Eugene-M]

First of all, thanks for the library.

I have an Intel Core Ultra 7 265K (with NPU) and two NVIDIA GeForce 4070 Ti Super GPUs on board.

• Currently, as far as I understood, it's impossible to configure library to run model with NPU enabled mode.

• At the same time I can't select GPU1 to load the model:

  My attempt to do this was not successful:
  const string cuda = "cuda";
  var config = new Config(gpuModelPath);
  config.ClearProviders();
  config.AppendProvider(cuda);
  config.SetProviderOption(cuda, "device_id", "0");
  config.SetProviderOption(cuda, "enable_cuda_graph", "1");
  var model = Model(config);

  If I change "device_id" from 0 to 1 I get the following error:
  Microsoft.ML.OnnxRuntimeGenAI.OnnxRuntimeGenAIException: 'Exception during initialization:
  D:\a_work\1\s\onnxruntime\core\providers\cuda\cuda_call.cc:129 onnxruntime::CudaCall
  D:\a_work\1\s\onnxruntime\core\providers\cuda\cuda_call.cc:121 onnxruntime::CudaCall CUDNN failure 2007:
  CUDNN_STATUS_BAD_PARAM_STREAM_MISMATCH ; GPU=1 ; hostname=<MY_HOST> ;
  file=D:\a_work\1\s\onnxruntime\core\providers\cuda\cuda_execution_provider.cc ; line=183 ; expr=cudnnSetStream(cudnn_handle_,
  stream);

• Along with the model itself, there are many files: general_config.json, merges.txt , tokenizer_config.json, vocab.json. It would be nice if you specified a way to transfer filePaths to the Config object, as well as the model name itself. Currently, we only define a folder, and somewhere in the library you have filenames (filePaths) hardcode. This leads to a lack of control and forces me to abandon the library and use Microsoft.ML.OnnxRuntime instead. But there aren't many good examples for the latter either, especially if I want to use a generative model.