fix: partially fix the bug of load_model

[Wanglongzhi2001]

Load_model use reflection to revive LayerArgs and then revive Layer.

TensorFlow.NET/src/TensorFlowNET.Keras/Utils/generic_utils.cs

Lines 64 to 69 in adeed05

	var argType = Assembly.Load("Tensorflow.Binding").GetType($"Tensorflow.Keras.ArgsDefinition.{class_name}Args");
	if(argType is null)
	{
	return null;
	}
	var deserializationMethod = typeof(JToken).GetMethods(BindingFlags.Instance | BindingFlags.Public)

So each layer's args must be the pattern of {class_name}Args, otherwise it cannot be revived.

[Wanglongzhi2001]

But there still exists some bug when use Merge Layer such as Add layer or Substract layer. When use Merge Layer, there will be some error when buildMap. So user arr better to not use Merge Layer at present.