Add proper type conversions from solidity contract --> python

[kylexqian]

Currently when we convert fixed-point numbers back into numpy values in the SDK we by-default just conver it to np.float32.

In the inference node we actually convert them to their proper value type based on the ONNX tensor already. I think we should do the same in the SDK.

This could cause issues for users if, for example, a user is expecting an int result but gets back a `float32. They would have to convert on their own side.

This would require changes in the Sequencer to include the ONNX Tensor type used for the data (Does this have privacy concerns?).

[kylexqian]

@oliverwalsh7 This is actually the same as the ONNX Tensor --> NP type task you did a while ago on the inference node.