Dask-pytorch-ddp

We have added a boolean flag called anon to the S3ImageFolder class to allow users to specify that an anonymous S3 connection (eg, a public S3 bucket without credentials) is being used. The behavior still defaults to False which retains the current behavior of using environment credentials to authenticate for S3 bucket access.

To make user experience easier and clearer, we're changing the name of our existing library. `dask-pytorch` is deprecated and may be used in the future for a more multifaceted package. Since our work is primarily focused on making the Distributed Data Parallel functionality of PyTorch work smoothly with Dask, we chose `dask-pytorch-ddp` for the new name.

Some bug fixes, including corrected links to documentation. Thanks to the commenters who have submitted issues for us!

Enables easier and smoother integration of Dask and PyTorch.

Features
- Custom data object class `S3ImageFolder` allows direct-from-s3 access to files for training/inference in pytorch
- `dispatch` module gives tools for distributing training tasks to cluster with minimal user fuss
- `results` module gives tools for managing the results and training statistics, for better model evaluation