Tensorflow-hub

- Enforce Python 3.5+ and TF1.15+.
- Add ModelLoadFormat setting to switch between (un)compressed model loading.
- Support for RaggedTensor inputs/outputs is backported from TF2 SavedModels to the deprecated hub.Module class (Use of tf.compat.v1.ragged.placeholder() in a module_fn is broken for TF2.3 and requires TF2.2 or TF2.4).
- Bug fixes.

* Add SavedModel.LoadOptions to hub.KerasLayer API to pass to load_v2.
* Improved error messaging.
* Documentation improvements.
* Bug fixes.

* Implemented make_nearest_neighbour_index tool.
* Added text FeatureColumn, hub.text_embedding_column_v2, for TF 2.x.
* CORD-19 embedding colab.
* Documentation improvements.
* Bug fixes.

* Added support for HubModule v1 in KerasLayer with default tags/signatures.
* Added support for the KerasLayer to specify tags, signature, as_dict, and
output_key arguments.
* Miscellaneous fixes to `hub.KerasLayer`
* Documentation update for TensorFlow 2.
* Use `hub.load()` and `hub.KerasLayer` with TF2 (also works in 1.15).
* For TF1, `hub.Module` and it's associated APIs remain available.

* Added two examples for exporting of Hub/TF2 modules.
* Switched to `dense_features_v2` if a high enough TF version is installed.
* Added `tools/make_image_classifier` for use with TF2.

* Fixes in `hub.KerasLayer` related with regularizers, config serialization
and usage with `tf.estimator`.
* Updates to feature columns to be compatible with feature columns V2 apis and
add `hub.sparse_text_embedding_column` for bag of words features.
* Made `hub.Module` usable within `tf.compat.v1.wrap_function`.