Coba 4.0.0 Pull Request

[mrucker]

This pull request upgrades contains a number of API changes and functionality upgrades.

At a very high level this pull request has three main goals:

1. Improve support for streaming large data sets
2. Make Experiment extensible for techniques such as off-policy evaluation
3. Refactor simulations to support more kinds of data (e.g., logged bandit data )

More specifically this pull request contains the following

• Separation of multiprocessing into Coba multiprocessing and Pipes multiprocessing
• Refactor of backend file processing to a pure streaming implementation
• Refactor of backend file processing to use a much simpler sparse representation
• Refactor of encoding module to have a simpler implementation and testing
• Refactor of CobaConfig to have a simpler and easier to use interface
• Upgraded DiskCacher implementation and interface so that data can be streamed into and out of it
• Upgraded HttpIO so that http requests can be streamed instead of downloaded all at once
• Added Resevoir algorithm to the Take filter for improved memory optimization when streaming
• Renamed simulations module to environments and added three types of environments: simulated, logged, warmstart
• Renamed benchmarks module to experiment (this is mostly for understandability reasons, not because of new functionality)
• Created a new Environments helper class that contains commonly used environment functionality