Similar to rubinius' .rbc files and python's .pyc files

Initial spike just uses a marshaled iseq, but once this is fleshed out we could switch to a raw format that is faster to dump/load. A header...