Why do src_key_padding_mask and src_mask have different conventions in TransformerEncoderLayer?

Question is pretty self-explanatory, seems like a potential source of bugs.