Pinning keras and tensorflow versions in setup

[lizgzil]

The new Keras version 2.4.0 released in June 2020 made changes that broke the deep reference parser.

 File "/usr/local/lib/python3.6/site-packages/deep_reference_parser/deep_reference_parser.py", line 989, in load_weights
    saving.load_weights_from_hdf5_group(

AttributeError: module 'keras.engine.saving' has no attribute 'load_weights_from_hdf5_group'

Thus I've pinned the version of keras to 2.2.5 (which is whats in the requirements). I also needed to pin the version of keras-contrib since it overwrote keras 2.2.5 to the newer version. Then I got issues with tensorflow, so also pinned this to the version in the requirements.

[codecov-commenter]

Codecov Report

Merging #39 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master      #39   +/-   ##
=======================================
  Coverage   81.46%   81.46%           
=======================================
  Files          17       17           
  Lines        1338     1338           
=======================================
  Hits         1090     1090           
  Misses        248      248           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 85949cb...51b5912. Read the comment docs.

[ivyleavedtoadflax]

Makes sense @lizgzil 👍 Long term i think it would be worth moving to tf.keras with tf 2.0, what prevented me in the past was that CRFs were not implemented, but it looks like they have been added now: tensorflow/addons#1999

[ivyleavedtoadflax]

Also referenced here: #28