Ml4ir

Changed

- Performing pre-processing step in `__init__()` to be able to copy files before model_config and feature_config are
initiated.

Changed

- Making pyspark an optional dependency to install ml4ir

Added

- Support for performing pre-processing steps (such as copying data) by custom class inheriting RelevancePipeline.

Added

- Support for using native tf/keras feature functions from the feature config YAML

Changed

- TFRecord format changed for SequenceExample to earlier implementation.
- Removed support for `max_len` attribute for SequenceExample features.
- No effective changes for Example TFRecords.
- TFRecord implementation on python (training) and jvm (inference) side are now in sync.

Added

- Changelog file to track version updates for ml4ir.
- `build-requirements.txt` with all python dependencies needed for developing on ml4ir and the CircleCI autobuilds.
- Updated CircleCI builds to use `build-requirements.txt`

Fixed

- Removed build requirements from the base ml4ir `requirements.txt` allowing us to keep the published whl file dependencies to be minimal.