Omniduct

This release exposes the new PySpark client by default through the Duct registry.

This release adds a database backend atop of PySpark.

This release changes the behaviour of AWS logins. In particular, `opinel` is only used if explicitly requested. This makes the behaviour of `omniduct` more predictable compared to direct use of `boto3`, etc.

This release stops passing `'default'` as the default value for `aws_profile` for S3 filesystems, as this breaks default behaviour for AWS logins.

Another relatively small release with minor breaking changes to the API, pushing it up to the next major version.

Breaking changes:
- All `async` keywords have been replaced with `wait`, in order to support Python 3.7.

Features and enhancements:
- Compatibility fixes for Python 2 data encoding in filesystem operations.
- S3 write support, as well as slightly more robust S3 client connections and the option to skip Hadoop artifacts.
- Improved logging and warnings (including support for indeterminate progress bars).
- Pyhive logging of query has been silenced.
- Added thrift dependency which does not seem to be pulled in upstream.

Behind the scenes:
- Stricter code conventions and linting has resulted in cleaner code.

This is a relatively small release, but which has slightly different default behaviour, making it a tick over to the next major release.

Improvements:
- Authorisation is now actually passed through to backends in Presto, Hive, Druid and Neo4j clients when `username` is not None. To avoid default usernames being passed, use `username=False` in the constructor.
- Allow customisation of the `source` name for Presto clients.

Bugfixes:
- Fixed `requests` being loaded at import time, despite supposedly being a runtime import.
- `None` results from presto polling are now correctly handled.