[WIP] Change parallelization to work by CLONE

[jchodera]

I'm not sure how fragile the current code is, but it seems to be working for now. The big questions are:

• Does termination in the middle of either trajectory concatenation or solvent stripping corrupt trajectories? So far, seems to work OK
• How many processes should we use in parallel? I've tested up to 16
• Why does this code produce large temporary files, and will they continue to grow, or do they get cleaned up along the way? Watching this now.

This currently switches munging to a new data directory.

We should set up nosetests and travis testing at some point.