Sending/receiving a numpy array containing NaN values to/from an mlserver with the NumpyCodec results in a ValueError: Out of range float values are not JSON compliant error. Is this a known limitation, and are there any good workarounds that would still allow me to use a pre-packaged server? I understand that it would probably be bad to assume than NaN can be mapped to null in the JSON, but maybe that could be an option?