Use of offpsring eventually leads to too many open files if there are closed shards

[jeffglover]

While doing some testing of kinesis-python I was evaluating how well it handles live resharding. As old shards remained kinesis-python would spin up a ShardReader and shut it down for each inactive shard. Offspring keeps track of these instances only to cleanup on exit, but they build up overtime and eventually resulting in [Errno 24] Too many open files

Offspring looks dead, but I submitted a pull request that would fix it. borgstrom/offspring#3