[Todo] Fix multi-input argument mapping

[whbldhwj]

For muliti-input model, when converting model to Relay graph, TVM may generate a different order of input arguments comapred to the original model.
For example, for the model below

class Q6Net(nn.Module):
    def __init__(self) -> None:
        super(Q6Net, self).__init__()

    def forward(self, data_discount, data_quantity, data_shipdate, data_extendedprice):
       ...

TVM-generated Relay graph takes the input arguments in the order of

[
Var(data_extendedprice, ty=TensorType([10], float32)), 
Var(data_discount, ty=TensorType([10], float32)), 
Var(data_quantity, ty=TensorType([10], float32)), 
Var(data_shipdate, ty=TensorType([10], float32))
]

This information can be extracted from:
https://github.com/apache/tvm/blob/2625878abef4bc78da65918a8a8c1db441638e8b/python/tvm/relay/frontend/pytorch.py#L4276

As a result, when executing the RAF model with the arguments in the original order
data_discount, data_quantity, data_shipdate, data_extendedprice
we will get the incorrect output due to the mismatch of arguments. This needs to be fixed.