Write corpus information to the experiment directory.

[oadams]

Figure out best how, and what else should get stored like utt2spk, prefixes, etc.

Some questions that spring to mind: Should we just pickle the corpus object and store it in exp/*? Should there be a human-readable format additionally?

What component is responsible for writing the corpus? Construction of the corpus object alone doesn't say anything about the experiment. model.py has been responsible for writing things to the experiment dir, but it seems like whatever code creates the exp_dir and feeds it to model.py should also be writing the Corpus object there.

[oadams]

The label inventory should also be written.

[shuttle1987]

If it is easy to do I'd definitely suggest that metadata is stored in a human readable serialization format, this makes it much easier to deal with things when they go wrong.

[shuttle1987]

I think this is related to #153 and #154