Support for Tensorflow 2.17

[mhuen]

Adds support for newest tensorflow version. This PR includes two updates:

• Convert KerasVariable to tensor
• fix eager execution for LR scheduler

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Please upload report for BASE (master@f2d0832). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff            @@
##             master     #11   +/-   ##
========================================
  Coverage          ?   0.06%           
========================================
  Files             ?      32           
  Lines             ?    2929           
  Branches          ?     534           
========================================
  Hits              ?       2           
  Misses            ?    2927           
  Partials          ?       0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mhuen]

Note: while falling back to compat.v1 optimizers works, support for variable learning rates is lost that was previously available. The issue is that newer tensorflow optimizers utilize keras optimizers which do not return the update operation when calling the apply_gradients method.
A proper solution is to convert everything to TF2 instead of relying on tf.compat.v1 symbols. See also #7 . This requires and extensive rewrite though.

[mhuen]

tf.compat.v1 symbols are now removed and everything is updated to proper tensorflow 2 logic. This fixes issue #7 .
The documentation has now also been updated to reflect these changes. This PR will thus also resolve issues #3 and #4.