Omniduct

A minor release that improves query times the `PySparkClient`, which formerly used `DataFrame.toLocalIterator()` instead of `.collect()` in non-streaming cases, resulting in slower and more memory consuming queries where streaming was not required.

This is a minor bugfix release fixing two issues:

- `DuctRegistry` configuration parsing broke when configuration contained empty dictionaries. (88 , thanks naoyak).
- Improved formatting of linebreaks in description text in the `show_about` tooling.

This release marks an important milestone: Omniduct is now considered to be stable and feature complete!

This release focussed primarily on improving documentation, and every user-facing class and function is now documented. There were some breaking changes to the `DuctRegistry` API which hitherto has not received much love, but otherwise code changes were fairly minimal.

Features and enhancements:
- Extended the `DuctRegistry` configuration format support, and made minor changes to the API to be more intuitive. [Users of omniduct v0.9.6 and earlier will need to adjust their use of `DuctRegistry` appropriately.
- Added a `show_about` utility function to simplify showing of information about the Omniduct package (or dependent packages), providing the option to "endorse" Omniduct or other significant dependencies.
- Included the new Omniduct logo, which you can feel free to use to endorse Omniduct in your own work.
- Extended documentation coverage to all user-facing classes and methods, and improved the consistency of documentation conventions.
- Added an example company/organisation-specific Omniduct configuration wrapper to assist with the bootstrapping of such wrappers.

Bugfixes:
- Avoid use of `_file_read` in `download` in case the `_open` method was overloaded by clients, as in `LocalFsClient`.

A minor release with two enhancements.

Features and enhancements:
- Add support for formatting the names of database query results (86)
- Allow rendering of parsed namespaces using overridden quotes and separators

A minor release with several feature enhancements.

Features and enhancements:
- Enable multi-value insert mode for `dataframe_to_table` methods (from upstream pandas patches not landed in master)
- Implement `mkdir` and `remove` for the HDFS filesystem client.
- Only check dependencies for the actual protocol (rather than all potential protocols) being used by the SQLAlchemy client.

Another minor release with several feature enhancements and bugfixes. Some minor breakage may occur if you overrode the default behaviour and use the `impyla` HiveServer2 backend (you will need to manually install the `impyla` dependency).

Features and enhancements:
- Snowflake protocol is supported via the sqlalchemy backend
- Database queries without result sets no longer result in exceptions
- `FileSystemClient.mkdir` methods now default to raising exceptions if path already exists (passing `exist_ok=True` will disable this behaviour)
- S3 `mkdir` and `remove` operations are now implemented

Bugfixes:
- Fixed `dataframe_to_table` functionality when using a remote CLI for HiveServer2 (the use of backtick/grave quotations in recent releases broke the scripted command)
- LocalFsClient used a Python 3 style `super()` call, breaking compatibility with Python 2
- LocalFsClient triggered preparation during initialisation, sometimes resulting in some spurious logging events
- Database queries without result sets no longer result in exceptions

Miscellaneous:
- `pyhive` is now the only HiveServer2 client installed by default (`impyla` does not support Python 3.7, but is otherwise still usable but must be manually installed).