Translation from ONNX type to MLIR type

[chentong319]

@tungld @tjingrant
This is related to PR#156.
We need a translation table from types in ONNX to types in MLIR. We can not do one-to-one mapping because some type is not supported by ONNX-MLIR or we may choose to support types in different way.
This translation will be used in Builder to build data for ONNX-MLIR from protbuf, and in gen_onnx-mlir.py for operation Tablegen.
How should we share the code to make sure we have a consistent translation?

[tungld]

One way to remove duplication is to define a static method in the op definition, say:

let extraClassDeclaration = [{                                                                                                                  
   static StringRef getTypeMapAttrName() {                                         
     return "to";                                                                  
   }                                                                               
}];

And, in FrontendDialectTransformer.cpp, we will obtain the attribute with that name, get the attribute value and pass it to convertONNXTypeToMLIRType to get an MLIR type, like:

// If the result type is specified by an attribute, set the result type        
// direcly.                
if (!op.getTypeMapAttrName().empty()) {                                        
  auto typeAttr = op.getTypeMapAttrName();                                     
  int typeNum = -1;                                                            
  for (int i = 0; i < node.attribute_size(); ++i) {                            
    auto attr = node.attribute(i);                                             
    if (attr.name() == typeAttr) {                                             
      typeNum = attr.i();                                                      
      break;                                                                   
    }                                                                          
  }                                                                            
  if (typeNum != -1) {                                                         
    auto resultType =                                                          
      mlir::UnrankedTensorType::get(convertONNXTypeToMLIRType(               
        static_cast<onnx::TensorProto_DataType>(typeNum)));                
    for (int i = 0; i < node.output().size(); i++) {                           
      if (variadicOut)                                                         
        (*(op.getODSResults(0).begin() + i)).setType(resultType);              
      else                                                                     
        (*op.getODSResults(i).begin()).setType(resultType);                    
    }                                                                          
  }                                                                            
}

It works well. However, this way does not use the static method getTypeMap nor resultTypeInference. Hence, it may make situation complicated and confusing.

[chentong319]

The design looks good to me.