Pre-trained/fixed embeddings for features

As far as I understand the lua code, pre_word_vecs/fix_word_vecs options are not used for the features part, and/or don’t have their equivalents for the features part. I would enjoy to get them in my experimentations with multiplexed input/output streams.