pluggable URI downloaders

[mlin]

The runtime's URI downloading code is meant to be extensible, by providing custom WDL tasks to download based on the URI scheme (s3://, gs://, etc.). The internal foundation is there (#260 #279), but we need to define and document the plug-in mechanism.

[mlin]

Useful: https://github.com/ActiveState/appdirs

[mlin]

We've put the heavy lifting for URI downloading into protocol-specific WDL tasks, which lets them benefit from all task runtime's built-in functionality (fetching docker images for binary dependencies, running them asynchronously, parallel scheduling & resource isolation, logging, error/signal handling, retry logic, ...).

Three further considerations for the downloader plug-in mechanism:

1. How are plugins installed & discovered by the runtime
2. How to provide any security credentials the downloader may need (e.g. AWS key id & secret)
3. Answers should generalize to other kinds of plugins we're going to have eventually

---

setuptools entry points are a strong candidate for how plugins should be installed and registered, then discovered with pkg_resources.

Other helpful:

https://amir.rachum.com/blog/2017/07/28/python-entry-points/ (blog)
https://entrypoints.readthedocs.io/en/latest/ (streamline discovery)
conda/conda#404 (conda considerations)
jupyter/notebook#2894 (discussion of numerous practical issues in the analogous problem for Jupyter)

@kislyuk

[mlin]

https://github.com/DropD/reentry