Causal interpretation rules for encoding and decoding models in neuroimaging